Explore the full ebook collection and download it now at ebookball.

com

Distributed and Shared Memory Algorithm for

Parallel Mining of Association Rules 1st Edition
by Hernández Palancar, Fraxedas Tormo, Festón
Cárdenas, Hernández León 9783540734987
https://ebookball.com/product/distributed-and-shared-memory-
algorithm-for-parallel-mining-of-association-rules-1st-
edition-by-herna-ndez-palancar-fraxedas-tormo-festa3n-ca-
rdenas-herna-ndez-lea3n-9783540734987-8926/

OR CLICK HERE

DOWLOAD EBOOK

Browse and Get More Ebook Downloads Instantly at https://ebookball.com

Click here to visit ebookball.com and download ebookball now
 Your digital treasures (PDF, ePub, MOBI) await
Download instantly and pick your perfect format...

Read anywhere, anytime, on any device!

Parallel and Distributed Programming Using C 1st Edition

by Cameron Hughes, Tracey Hughes ISBN 0321544676
9780321544674
https://ebookball.com/product/parallel-and-distributed-programming-
using-c-1st-edition-by-cameron-hughes-tracey-hughes-
isbn-0321544676-9780321544674-12626/

ebookball.com

LNCS 2834 Parallel Algorithm for Mining Maximal Frequent

Patterns 1st edition by Hui Wang, Zhiting Xiao, Hongjun
Zhang, Shengyi Jiang ISBN 3540200541 978-3540200543
https://ebookball.com/product/lncs-2834-parallel-algorithm-for-mining-
maximal-frequent-patterns-1st-edition-by-hui-wang-zhiting-xiao-
hongjun-zhang-shengyi-jiang-isbn-3540200541-978-3540200543-9208/

ebookball.com

An Empirical Comparison of Ideal and Empirical ROC Based

Reject Rules 1st Edition by Claudio Marrocco, Mario
Molinara, Francesco Tortorella 9783540734987
https://ebookball.com/product/an-empirical-comparison-of-ideal-and-
empirical-roc-based-reject-rules-1st-edition-by-claudio-marrocco-
mario-molinara-francesco-tortorella-9783540734987-11798/

ebookball.com

Blog Mining for the Fortune 500 1st Edition by James

Geller, Sapankumar Parikh, Sriram Krishnan 9783540734987

https://ebookball.com/product/blog-mining-for-the-fortune-500-1st-
edition-by-james-geller-sapankumar-parikh-sriram-
krishnan-9783540734987-8796/

ebookball.com
Algorithms and Protocols for Wireless Sensor Networks
Wiley Series on Parallel and Distributed Computing 1st
Edition by Azzedine Boukerche 0471798134 978-0470396353
https://ebookball.com/product/algorithms-and-protocols-for-wireless-
sensor-networks-wiley-series-on-parallel-and-distributed-
computing-1st-edition-by-azzedine-
boukerche-0471798134-978-0470396353-13402/
ebookball.com

Association Rule Hiding for Data Mining 2010th Edition by

Aris Gkoulalas Divanis, Vassilios Verykios ISBN 1441965688
9781441965684
https://ebookball.com/product/association-rule-hiding-for-data-
mining-2010th-edition-by-aris-gkoulalas-divanis-vassilios-verykios-
isbn-1441965688-9781441965684-13572/

ebookball.com

Advances in Parallel Distributed Computing First

International Conference on Parallel Distributed Computing
Technologies and Applications PDCTA 2011 Tirunelveli Tamil
Nadu India September 23 25 2011 Proceedings 1st edition by
https://ebookball.com/product/advances-in-parallel-distributed-
Dhinaharan Nagamalai, Eric Renault, Murugan Dhanuskodi
computing-first-international-conference-on-parallel-distributed-
computing-technologies-and-applications-pdcta-2011-tirunelveli-tamil-
3642240364 9783642240362
nadu-india-september-23-25-2011-p/
ebookball.com

Memory as a Programming Concept in C and C 1st Edition by

Frantisek Franek isbn 0521520436 9780521520430

https://ebookball.com/product/memory-as-a-programming-concept-in-c-
and-c-1st-edition-by-frantisek-franek-
isbn-0521520436-9780521520430-12552/

ebookball.com

Memory as a Programming Concept in C and C 1st Edition by

Frantisek Franek ISBN 051183862X 9780511838620

https://ebookball.com/product/memory-as-a-programming-concept-in-c-
and-c-1st-edition-by-frantisek-franek-
isbn-051183862x-9780511838620-14982/

ebookball.com
 Distributed and Shared Memory Algorithm for
Parallel Mining of Association Rules

J. Hernández Palancar, O. Fraxedas Tormo, J. Festón Cárdenas,

and R. Hernández León

Advanced Technologies Application Center (CENATAV), 7a  21812 e/ 218 y 222,

Rpto. Siboney, Playa, C.P. 12200, La Habana, Cuba
{jpalancar,ofraxedas,jfeston,rhernandez}@cenatav.co.cu

Abstract. The search for frequent patterns in transactional databases

is considered one of the most important data mining problems. Several
parallel and sequential algorithms have been proposed in the literature to
solve this problem. Almost all of these algorithms make repeated passes
over the dataset to determine the set of frequent itemsets, thus imply-
ing high I/O overhead. In the parallel case, most algorithms perform a
sum-reduction at the end of each pass to construct the global counts,
also implying high synchronization cost. We present a novel algorithm
that exploits efficiently the trade-offs between computation, communi-
cation, memory usage and synchronization. The algorithm was imple-
mented over a cluster of SMP nodes combining distributed and shared
memory paradigms. This paper presents the results of our algorithm on
different data sizes experimented on different numbers of processors, and
studies the effect of these variations on the overall performance.

1 Introduction
The discovery of Association Rules is one of the most productive fields in the
development of sequential algorithms as well as parallel algorithms for Data
Mining. Simultaneously, with the evolution of these algorithms the possible ap-
plications of Association Rule Mining (ARM) has also been extended together
with a corresponding increase in the volume of the databases to be mined. As
a consequence of the latter, even using the most efficient sequential ARM al-
gorithms, it is not possible to reduce the support threshold to the desired level
without causing a combinatorial explosion in the number of identified frequent
itemsets coupled with a corresponding computational overhead.
The situation described above confirms the relevance of the application of
Parallel Computing for Association Rule Mining, which is a very active global
research area. The main challenges of Parallel Computing are: load balancing,
minimization of the inter-process communication overhead, the reduction of syn-
chronization requirements and effective use of the memory available to each
processor.
These issues must be taken into account in the development of efficient parallel
algorithms for Association Rule Mining; basic references to consider are [8,9,10].

P. Perner (Ed.): MLDM 2007, LNAI 4571, pp. 349–363, 2007.


c Springer-Verlag Berlin Heidelberg 2007
350 J.H. Palancar et al.

The prototypical ARM application is the analysis of sales or basket data [1].
The task can be broken into two steps. The first step consists of finding the set of
all frequent sets of items that can be the transaction database. The second step
consists of forming implication rules among the sets of items found; the latter
can be done in a straightforward manner so we will focus on the first step.
In previous papers [11,12], we proposed a new algorithm called CBMine (Com-
pressed Binary Mine) for mining association rules and frequent itemsets. Its ef-
ficiency is based on a compressed vertical binary representation of the database.
CBMine has been compared with other efficient ARM algorithms to obtain fre-
quent itemsets, including: Fp-growth (implementation of Bodon), MAFIA and
Patricia Trie . The experimental results obtained showed that CBMine gives the
best performance in most cases, especially on big and sparse databases.
In this paper we propose a new parallel algorithm based in CBMine named
ParCBMine(Parallel Compressed Binary Mine). ParCBMine exploits efficiently
the trade-offs between computation, communication, memory usage and syn-
chronization. The algorithm was implemented over a cluster of SMP nodes com-
bining distributed and shared memory paradigms. Section 5 of this paper shows
the experimental results of our algorithm on different data sizes, evaluated on
different numbers of processors, and studies about the effect of these variations
on the overall performance.
The paper is organized as follows: the next section is dedicated to related work;
in section 3 we give a formal definition of association rules; section 4 contains
a description of ParCBMine algorithm; experimental results are discussed in
section 5; and some conclusions are presented in section 6.

2 Related Work

Until now the great majority of the parallel algorithms for Association Rule
Mining are based on the sequential Apriori algorithm. An excellent survey made
by Zaki in 1999 [18] classifies different algorithms up to that date, according to
the load balance strategy, the architecture and the type of parallelism used in
the algorithm. Other important references are [2,15,16,19,20,22].
Apriori algorithm has been the most significant of all sequential algorithms
proposed in the literature. Yet, directly adapting an Apriori-like algorithm will
not significantly improve performance over frequent itemsets generation. To per-
form better than Apriori-like algorithms, we must focus on the disadvantages as-
sociated with this approach. The main challenges include synchronization, com-
munication minimization, work-load balancing, finding good data layout and
data decomposition, and disk I/O minimization.
Recently interesting parallel ARM has increased as a result of this early work.
We can identify a number of early ARM algorithms: Count Distribution, Data
Distribution and Candidate Distribution. These algorithms were first presented
in [2] and offer a fairly simple parallelization of Apriori using different paradigms
of parallelization; namely data-parallelism and control-parallelism, or a combi-
nation of both. In Count Distribution the dataset is partitioned equally among
 Distributed and Shared Memory Algorithm 351

the nodes of the parallel system. Each of these nodes computes the local support
for every candidate k-itemset in the iteration k. At the end of each iteration by
exchanging the local supports the global support is generated and the frequent
itemsets determined. The nodes must be synchronized to receive the candidate
itemsets and the coordinator node must wait for all local counts to generate
the global support. The former factors affects communication cost and load-
balancing; however the Count Distribution algorithm represents a good first
step and can be the core of subsequent implementations that address these is-
sues. In Data Distribution the set of candidate itemset is partitioned into disjoint
sets and these are sent to different nodes. The problem in this parallel version
of Apriori is the magnitude of the huge communications required at the end of
each iteration. In Candidate Distribution load-balancing is thus the main target,
selectively replicating the dataset so that each processor proceeds independently.
The algorithm requires redistribution of the dataset at level l, this is identified
using a heuristic approach.
There are other parallel versions of well-known sequential algorithms like PDM
(parallelizing DHP) [3]. But this was not a successful attempt due to its poor
performance with respect to the above algorithms. Other algorithms that ad-
dress the size of candidacy and better pruning techniques are DMA and FDM
presented in [4,5]. In [7] the Optimized Distributed Association Mining (ODAM)
algorithm is proposed based on Count Distribution which reduce both the size
of the average transaction and the number of message exchanges among nodes
in order to achieve better performance.
The Eclat(Equivalence CLass Transformation) algorithm [17] uses an itemset
grouping scheme based on equivalence classes and partitions them into disjoint
subsets among the processors. At the same time Eclat makes use of a kind of
vertical representation of the dataset and then selectively replicates it so that
each processor has the portion of the dataset it needs for calculations. After
the initial phase the algorithm eliminates the need for later communication or
synchronization. The algorithm scans the local partition of the dataset three
times, therefore diminishing the I/O overhead. Unlike other earlier algorithms,
Eclat uses simple intersection operations to compute frequent itemsets and does
not use complex hash tables structures. The main deficiency of this algorithm
lies in the need for a proper heuristic to achieve a suitable load balance among
the processors as of the L2 partitioning, because the equivalence classes do not
have the same cardinality.
In [15] a collection of algorithms with different partitioning and candidate
itemsets count schemes are described. Like Eclat, all of them assume a vertical
representation of the dataset (tidlits per item), which facilitates the intersection
operation of tids of items that make up an itemset. The dataset is duplicated
in a selective fashion to reduce synchronization. Two of these algorithms (Par-
Eclat and Par-MaxEclat) are based on the classes of equivalence formed by the
candidates first item, whereas the other two algorithms (Par-Clique and Par-
MaxClique) use the maximum closed hypergraph to partition the candidates.
352 J.H. Palancar et al.

In [21], a parallel algorithm is proposed for Association Rule Mining that uses
a classification hierarchy named HPGM (Hierarchical Hash Partitioned Gener-
alized Association Rule Mining). In this algorithm, the available memory space
is completely used identifying the frequent occurrence of candidates itemsets
and replicating them to all processors, considering that frequent itemsets can be
locally processed without communication. This way the load asymmetry among
processors can be effectively reduced.

3 Problem Definition
In this section we define some necessary terminology to facilitate understanding
of the following sections. In this context it should be noted that we are only
focused on the problem of identifying frequent itemsets on large databases.
A dataset is a set of transactions and each of these is composed by a transac-
tion identifier (TID) and a set of items. The items in a transaction may represent
a shopping list in a supermarket by a customer (known as basket data) or words
in a document or stocks movements. A set of items, called itemset is frequent
if it is contained in a number of transactions above a user-specified threshold
(minimum support-minsup).
An itemset with k items will be referred to as k-itemset and its support will be
denoted as X.sup, where X is the k-itemset in question; support is represented
as a percentage rather than an absolute number of transactions.
More formally: I = {i1 , i2 , ..., in } be a set of n distinct items. Each transaction
T in the dataset D contains a set of items, such that T ⊆ I. An itemset is said
to have a support s if s% of the transaction in D contains the itemset.

4 ParCBMine
In this section we describe the parallel version of the CBMine algorithm which
we have named ParCBMine (Parallel CBMine).
ParCBMine takes advantage of the vertical representation of the dataset as in
CBMine and combines suitably the parallel programming models of shared and
distributed memory using the libraries pthreads (for multithreads programming)
and MPI (for message passing programming) respectively.
The mixture of multithreads programming and message passing in ParCB-
Mine was done based on the fact that the algorithm was implemented over a
cluster with SMP (Symmetric Multi-Processing) nodes for the parallel process-
ing managed by a GNU/Linux operating system; each node is composed by a
dual processor. All processors are Intel Xeon with hyperthreading technology,
which provides up to four threads on each node.
Although the algorithm is not tied to the number of real threads (proces-
sors) that could be deployed on each node, this is an important element on the
scalability of ParCBMine, because it allows the use of global information in the
shared memory of each node in a better way, i.e., in candidate generation and
support counting. Many authors refer to this as “intra-node parallelism”, and in
 Distributed and Shared Memory Algorithm 353

certain way we have incorporated some aspects of the Candidate Distribution

algorithm, in this case by means of multithread programming using the Pthreads
library.

4.1 CBMine Algorithm

CBMine is a breadth-first search algorithm with a VTV organization, that uses

compressed integer-lists for itemset representation.
Let T be the binary representation of a database, with n filtered items and m
transactions. Taking from T the columns associated with frequent items, each
item j can be represented as a list Ij of integers (integer-list) of word size w, as
follows:
Ij = {W1,j , . . . , Wq.j } , q = m/w, (1)

where each integer of the list can be defined as:


min(w,m−(s−1)∗w)
Ws,j = 2(w−r) ∗ t((s−1)∗w+r),j . (2)
r=1

The upper expression min(w, m − (s − 1) ∗ w) is included to consider the case

in which the transaction number (s − 1) ∗ w + r does not exist due to the fact that
it is greater than m. The value ti,j is the bit value of term j in the transaction i.
This algorithm iteratively generates a prefix list P Lk . The elements of this
list have the format: P ref ixk−1 , CAP ref ixk−1 , Suf f ixesP ref ixk−1 , where
P ref ixk−1 is a (k − 1)-itemset, CAP ref ixk−1 is the corresponding compressed
integer-list, and Suf f ixesP ref ixk−1 is the set of all suffix items j of k-itemsets
extended with the same P ref ixk−1 , where j is lexicographically greater than
every item in the prefix and the extended k-itemsets are frequent. This repre-
sentation not only reduces the required memory space to store the integer-lists
but also eliminates the Join step described in Apriori algorithm.
The Prune step of Apriori algorithm is optimized by generating P Lk as a
sorted list according to the prefix field and, for each element, by the suffix field.
In order to determine the support of an itemset with a compressed integer-list
CA, the following expression is considered:

Support(CA) = BitCount(Bs ), (3)
s,Bs ∈CA

where BitCount(Bs ) represents a function that calculates the Hamming Weight

of each Bs .
Although this algorithm uses compressed integer-lists of non null integers
(CA) for itemset representation, in order to improve the efficiency, we maintain
the initial integer-lists (including the null integers) Ij = {W1,j , . . . , Wq,j } asso-
ciated with each large 1-itemset j. This consideration allows direct accessing for
any Ij the integer position defined in CA.
354 J.H. Palancar et al.

The above allows us to define the following formula (notice that this func-
tion represents a significant difference and improvement with respect to other
methods):

CompAnd(CA, Ij ) = {s, Bs |s, Bs  ∈ CA, Bs = (Bs andWs,j ), Bs = 0}. (4)

Note that the cardinality of CA is reduced as the size of the itemsets increases
due to the downward closure property; thus the application of identities 3 and 4
becomes more efficient.
The complete CBMine algorithm is presented in Table 1.

Table 1. CBMine algorithm

Algorithm 1: CBMine
1 L1 = {large 1-itemsets} ; // Scanning the database
2 P L2 = {P ref ix1 , CAP ref ix1 , Suf f ixesP ref ix1 };
3 for k = 3;P Lk−1 = ∅; k + + do
4 forall P ref ix, CA, Suf f ixes ∈ P Lk−1 do
5 forall item j ∈ Suf f ixes do
6 P ref ix = P ref ix ∪ {j};
7 CA = CompactAnd(CA,Ij );
8 forall (j  ∈ Suf f ixes) and (j  > j) do
9 if Prune(P ref ix ∪ {j  }, P Lk−1 ) and Support(CompactAnd(CA ,
Ij  )) ≥ minsup then
10 Suf f ixes = Suf f ixes ∪ {j  };
11 end
12 if Suf f ixes = ∅ then
13 P Lk = P Lk ∪ {P ref ix , CA , Suf f ixes };
14 end
15 end
16 end
17 end
18 end 
19 Answer= k Lk ; // Lk is obtained from P Lk

Note that this algorithm only scans the dataset once in the first step.

4.2 Intelligent Block Partitioning

Given P Lk−1 we need to partition it among the threads in the most efficient
manner. In the literature we can identified several partitioning techniques, such
as Bitonic Partitioning from Zaki [19]. Nevertheless, given the features of the
sequential algorithm, we achieve the best results making a block partitioning, so
the information to be processed by each thread was not fragmented.
For this purpose we develop Intelligent Block Partitioning (IBP), dynamically
recomputing the load balance for each thread in a straightforward manner. We
 Distributed and Shared Memory Algorithm 355

use equation 5 to compute the work load generated by a Prefix based on the size
of its Suffixes. The pseudo-code of IBP is given in Table 2.
x (x − 1)
G (x) = (5)
2

Table 2. Intelligent Block Partitioning algorithm

Algorithm 2: IBP
|P Lk−1 |
1 Total = j=0 G(|Suf f ixesP ref ixj |); /* Total Work Load */
2
Total
Ideal = M axT hreads ; /* Ideal Work Load for each thread */
3 i = 1;
4 load = starts[0] = 0;
5 forall P ref ix, CA, Suf f ixes ∈ P Lk−1 do
6 load = load + G(|Suf f ixes|);
7 if load > Ideal then
/* Set block boundaries */
8 start[i] = stop[i − 1] = P ref ix, CA, Suf f ixes;
/* Dynamically recompute Total and Ideal Work Load */
9 Total = Total- load + G(|Suf f ixes|);
10 Ideal = Total ;
M axT hreads−i
11 load = G(|Suf f ixes|);
12 i++;
13 end
14 end
15 stop[i] = P ref ix|P Lk−1 | , CAP ref ix|P L , Suf f ixesP ref ix|P L ;
k−1 | k−1 |

The aforementioned partition strategy is one of the improvements ParCB-

Mine introduces over its sequential counterpart, and this can be verified in the
experimental results.

4.3 ParCBMine Algorithm

Considering a master-slave framework, typical of parallel clusters, in the first
pass, the master node or coordinator determines the global L1 and partitions
the dataset D in N equitable segments and sends each one of them to the
corresponding node that makes up the cluster, of this way ParCBMine like Count
Distribution, adopts a horizontal partitioning of the dataset thus using “inter-
node parallelism”, in this case the communication among the nodes is made by
means of message passing using the MPI library.
The first pass is special. For all other passes k > 1, the algorithm works as
follows:
1. Each master-thread process Pj (j = 1, N ) generates all the set Ck , using
all the frequent itemsets Lk−1 created at the end of pass k − 1. Notice
356 J.H. Palancar et al.

that every process has the same Lk−1 , so they will generate identical Ck .
Threads Pi (i = 1, N × M axT hreads) running in the same node, share the
same memory structure for Lk−1 , Ck and Dj (j = 1, N ).
2. The master-thread process Pj creates M axT hreads−1 new threads and each
one of these makes a pass over Dj data partition and develops local support
counts for a portion of the candidates in Ck which was previously partitioned
using the IBP strategy. With this, the local Ck at node j, is partitioned
equitably and each thread of the process develops the support count of its
candidates without making any synchronization to access the memory, since
the support count is developed on a reserved memory structure for each
candidate, taking advantage of the vertical representation of the dataset.
3. The master-thread process Pj sends the local counts of Ck to the master
node or coordinator, in order to make an all-reduce operation to generate
the global counts of Ck . Master-thread processes are forced to synchronize
in this step.
4. The master node or coordinator computes Lk from Ck . If Lk is not empty
the coordinator sends it to the master-thread process Pj and continues on
to the next pass.

Notice that unlike Count Distribution we have replaced the word processor by
process, since given the characteristics of the hardware of our cluster the amount
of processes is greater than the amount of processors, and can be expressed by
N × M axT hreads, where: N is the number of nodes and M axT hreads is the
maximum number of threads per node, in our particular case M axT hreads = 4
considering the use of the Hyperthreading technology.
Unlike PAR-DCI algorithm [13], in which the local dataset is partitioned yet
again into as many portions as threads that were possible to deploy, in step 2 of
ParCBMine a more efficient solution was adopted. We distribute the candidate
support count among the threads by partitioning the candidate set into disjoint
parts of approximately the same size, without the need for semaphoric operations
to control memory access.

4.4 Complexity Analysis

In this section we will evaluate the complexity of our algorithm in three different
contexts, first assuming the use of shared memory model, second employing the
distributed memory model, and lastly the solution proposed by us of fusing the
models of shared memory and distributed memory.
Given that our algorithm is intended for a parallel framework based on an
SMP cluster, it is important to indicate that if we had used only a distributed
memory model based on message passing, like other authors have done, the
performance of the algorithm would have suffered considerably.
It is well known that in any algorithm based on Count Distribution the scala-
bility degrades as the number of dataset partitions increases, due to the amount
of information that each MPI process receives in each pass when synchroniz-
ing the processes in order to develop global support counts of itemsets in the
 Distributed and Shared Memory Algorithm 357

candidate set. The amount of information is (N − 1) × |Ck |, where: N is the

number of MPI processes with a dataset portion assigned to it and |Ck | is the
cardinality of the candidate set generated in each pass and for which each MPI
process computes a local support count. Bear in mind that whatever the value of
N is, the cardinality of Ck does not change. This is the reason why the reduction
of local set Ck is an issue that has been and continues to be a research objec-
tive. From the literature one suggested approach involves the use of probabilistic
estimations of local support, see [6,14].
Given that we are using SMP nodes the advantages offered by the data local-
ity would be wasted since all the MPI processes running in each node will try
to equitably distribute the total physical memory of the node, reserving equal
amounts of memory for data structures to store Lk−1 and Ck , as well as for the
dataset partition assigned to the node. For the analysis lets assume that the
problem size remains constant, so the amount of candidates will be the same in
each case and will be denoted as |Ck |.
In the development of a parallel algorithm the most common notation for the
execution times are (if we consider a problem of size m running in p processors):
Sequential computation denoted by σ (m), Parallel execution time (computa-
tion that can be performed in parallel) denoted by ϕ (m) and Parallel overhead
(communication and synchronization, etc) denoted by κ (m). For the experi-
ments performed, the sequential plus the parallel execution time was considered
as the parallel execution time because of the characteristic of the CBMine, the
part that can not be parallelized is less than 1% of total execution time. For that
reason, the two times measured were: Parallel execution time and Parallel
overhead.

Shared Memory: Lk−1 is partitioned in disjoint sets using IBP, support is

develop from the common data base.

Algorithm 3: Shared Memory

while(Lk−1 = ∅) ;  /* Level Iterator */
Ckt = IBP (Lk−1 , t) ; /* Ck = Ckt t=1,..,MaxThreads */
f oreach(X ∈ Ckt ) ; /* Count
 each X in the DB */
if (sup(X, DB) ≥ minsup) Lk = Lk {X}

where: ϕ(m) = |Ck | ∗ |DB|, κ(m) = ∅.

 |Ck | ∗ |DB| 
ShT ime = = ... = max|Ckt | ∗ |DB| (6)
p
k k

Distributed Memory: The DB is partitioned among the quantity process de-

noted by P , each one has a copy of Lk−1 , count the local support and
exchange it (i.e. Message Passing). For example: if the nodes are single pro-
cessor P = N ; in case that the nodes are SMP then P = N x p, where: N is
the quantity nodes, and p is the number of processors in each node.
358 J.H. Palancar et al.

Algorithm 4: Distributed Memory

DBid = P artition(DB, id) ; /* Horizontal partitioning */
while(Lk−1 = ∅) ; /* Level Iterator */
Ck = GenerateCandidate(Lk−1 ) ; /* Ck is the same for each
process */
f oreach(X ∈ Ck ) ; /* Count each X in the DBid */
local[x] = sup(X, DBid ) ; /* Local support */
global = InterchangeAndSum(local) ; /* All to All */
f oreach(X ∈ Ck ) ;  /* global support */
if (global[X] ≥ minsup) Lk = Lk {X}

 
where:
 ϕ(m) = |Ck |∗|DB|, κ(m) = k InterchangeAndSum = k |Ck |∗2∗P =
k |Ck | ∗ 2 ∗ N ∗ p.

 |Ck | ∗ |DB| 
DsT ime = + κ(m) = |Ck | ∗ max|DBid | + κ(m) (7)
N ∗p
k k

Share + Distributed Memory Solution (Hybrid memory): In the pre-

vious cases the P processes were sharing the memory or completely dis-
tributed, in this case there will be N MPI-processes in correspondence with
the quantity of nodes and in each node p processes sharing memory, for that
reason P = N , because for the communication among nodes is not consider-
ing the quantity of processes in each node. In this case a process master for
each node is in charge of the communication with the remaining nodes and
of distributing tasks to the other processes that are in its node.

Algorithm 5: Hybrid Memory

DBN = P artition(DB, N ) ; /* Horizontal partitioning */
while(Lk−1 = ∅) ;  /* Level Iterator */
Ckt = IBP (Lk−1 , t) ; /* Ck = Ckt t=1,..,MaxThreads */
f oreach(X ∈ Ckt ) ; /* Count each X in the DBN */
local[x] = sup(X, DBN ) ; /* Local support */
if (master(t)) then global = InterchangeAndSum(local)
f oreach(X ∈ Ckt ) ;  /* global support */
if (global[X] ≥ minsup) Lk = Lk {X}

 
where: ϕ(m) = |Ck |∗ |DB|, κ(m) = k InterchangeAndSum = k |Ck |∗ 2 ∗ N .

 |Ck | ∗ |DB| 
HyT ime = + κ(m) = max|Ckt | ∗ max|DBN | + κ(m) (8)
N ∗p
k k
 Distributed and Shared Memory Algorithm 359

If we are using the same processes quantity for each memory model it is very
simple to observe that:

ShT ime < HyT ime < DsT ime (9)

5 Results

All the experiments described in this section were performed on a SMP clus-
ter of which we used 6 nodes: the master node and five working nodes. Each
working node is equipped with two Intel Xeon processors at 2.4 GHz based on
hyperthreading technology, 512 MB of RAM, 40 GB of disk space and 1 Gb/s
Fast Ethernet card. The working nodes are connected to a master node by a
network switch Gigabit Ethernet. The master node is equipped with two Intel
Xeon processors at 3.06 GHz based on hyperthreading technology too, 2 GB of
RAM, and a disk array of five disks, 36.4 GB of disk space each (total 145.6 GB).
We ran two versions of the parallel algorithm: one using the distributed mem-
ory model implemented with MPI, so there were two processes for each node, i.e.
one process by physical CPU; and the other combining the distributed memory
model (MPI again) and the shared memory model implemented using Pthreads,
in this case there were 4 threads per node sharing the same memory, considering
the use of hyperthreading technology.
The experiments were made with one synthetic (T40I10D600K composed by
600000 transactions and 999 items) and one textual dataset (Kosarak com-
posed by 990007 transactions and 41935 items) (available from FIMI repository-
http://fimi.cs.helsinki.fi). The Kosarak Dataset was provided by Ferenc Bodon
and contains (anonymized) click-stream data of a Hungarian on-line news portal.
The T40I10D600K was created using an IBM generator(www.almaden.ibm.com/
cs/quest/syndata.html).
In the first experiment we compared the execution times between ParCBMine
using MPI plus Threads and ParCBMine using MPI only. The Figure 1 (a) and
Figure 1 (b) show that the parallel execution time is reduced to half when the
number of processors is doubled, for both implementations. The communication
time overhead is stable in the first case (with the use of MPI + Tthreads) but
increases linearly in the second (MPI only).
The second experiment was performed to analyze the SpeedUp (Figure 2)
and Efficiency (Figure 3) of both implementations of ParCBMine algorithm.
Figure 2 shows that the implementation of ParCBMine algorithm using MPI plus
Threads scales better when the number of processors is increased in spite of the
communication time overhead. Likewise, notice that in Figure 3 the degradation
of the efficiency for ParCBMine implementation using MPI plus Threads is much
slower with the increase of the communication time overhead.
In the third experiment we analyzed the algorithm scalability, thus we consid-
ered the case where both datasets were so big that they could not fit in the main
memory of any node, increasing databases size in proportion with the number
of nodes (N ), these datasets were named T40I10D600KxN and KosarakxN. The
360 J.H. Palancar et al.

(a) (b)

Fig. 1. Execution time comparison: (a) T40I10D600K, minsup = 0.01, (b) Kosarak,
minsup = 0.002

(a) T40I10D600K (b) Kosarak

Fig. 2. SpeedUp comparison

(c) T40I10D600K (d) Kosarak

Fig. 3. Efficiency comparison

minimum support values used in each case were the smallest that the sequential
version could process (for T40I10D600KxN the minimum support was set to
0.005 and for KosarakxN the minimum support was set to 0.003).
 Distributed and Shared Memory Algorithm 361

(a) MPI + Threads (b) MPI

(c) MPI + Threads (d) MPI

Fig. 4. Scalability Analysis of ParCBMine in T40I10D600KxN (a y b) and

KosarakxN(c y d)datasets

Observe in Figure 4 that the parallel execution time remains constant, thus
the scalability of the algorithm does not depend on the database used.
As a conclusion of these experiments we can affirm that the shared-distributed
memory combination proved to be an effective way to avoid high traffic of data
and drastic reduction of the efficiency of the parallel algorithm.

6 Conclusions

The algorithms proposed by Rakesh Agrawal and John Shafer in [2] are recog-
nized as benchmarks for the development of parallel algorithms for Association
Rules Mining.
Making a general assessment of these algorithms we can say that the Count
Distribution reduces the communication overhead at the expense of ignoring
the system physical memory. In a cluster of workstations environment, with
monoprocessor nodes, this is probably the best approach; nevertheless it may not
be the best solution in the case where nodes are SMP, because it would not take
advantage of the combination of shared and distributed memory models. In order
to reach efficient implementations based on Count Distribution, determining new
362 J.H. Palancar et al.

heuristics that allow a reduction of the cardinality of the local Ck obtained by

each processor continues to be a latent problem.
The Data Distribution algorithm can help us to explore this feature by fully
exploiting the physical memory with the risk of increasing communication over-
head. The ability to count in a single pass T times as many candidates as Count
Distribution makes this algorithm a strong contender.
If we include detailed background knowledge of the problem in the Candidate
Distribution, the joint benefits of Count Distribution and Data Distribution [2]
can be obtained. Yet, there are still some challenges for researchers in parallel
algorithms for association rule mining: to find a heuristic that allows (from a step
k = l) candidate itemsets partitioning so that synchronization among processors
is not needed, and to obtain a suitable load balance among the processors.
In conclusion we suggest that the purposed parallel algorithm described here
ParCBMine based on the sequential algorithm CBMine, and sustained on the
principles of Count Distribution in which some features of Candidate Distribu-
tion are also introduced, suitably combines the parallel programming based on
the message passing model with multithread programming. ParCBMine contin-
ues to be developed and in the future we expect to present new results oriented
to the reduction of the computational effort at synchronization level and to reach
a better load balance among processors.

References
1. Agrawal, R., Mannila, H., Srikant, R., Toivonen, H., Verkamo, A.I.: Fast discovery
of association rules. In: Fayyad, U., et al. (eds.) Advances in Knowledge Discovery
and Data Mining, MIT Press, Cambridge (1996)
2. Agrawal, R., Shafer, J.: Parallel mining of association rules. In IEEE Trans. on
Knowledge and Data Engg. 8(6), 962–969 (1996)
3. Park, J.S., Chen, M., Yu, P.S.: Efficient parallel data mining for association rules.
In: ACM Intl. Conf. Information and Knowledge Management (November 1995)
4. Cheung, D., Ng, V., Fu, A., Fu, Y.: Efficient mining of association rules in dis-
tributed databases. In IEEE Trans. on Knowledge and Data Engg. 8(6), 911–922
(1996)
5. Cheung, D., Han, J., Ng, V., Fu, A., Fu, Y.: A fast distributed algorithm for
mining association rules. In: 4th Intl. Conf. Parallel and Distributed Info. Systems
(December 1996)
6. Cheung, D.W., Xiao, Y.: Effect of Data Skewness in Parallel Mining of Association
Rules. In: Proceedings of the 2nd Pacific-Asia Conference on Knowledge Discovery
and Data Mining, pp. 48–60, Melbourne, Australia (April 1998)
7. Ashrafi, M.Z., Taniar, D., Smith, K.A.: ODAM: An Optimized Distributed Asso-
ciation Rule Mining Algorithm. IEEE Distributed Systems Online (5) (2004)
8. Freitas, A.A., Lavington, S.H.: Mining Very Large Databases with Parallel Pro-
cessing. Kluwer Academic Publishers, Boston (1998)
9. Freitas, A.A.: A Survey of Parallel Data Mining. In: Proc. 2nd Int. Conf. on the
Practical Applications of Knowledge Discovery and Data Mining (1998)
10. Skillicorn, D.: Parallel Data Mining, Department of Computing and Information
Science Queen’s University, Kingston (1999)
 Distributed and Shared Memory Algorithm 363

11. Palancar, J.H., León, R.H., Pagola, J.M., Dı́az, A.H.: Mining Frequent Patterns
Using Compressed Vertical Binary Representations In: Lin, T.Y., Xie, Y. (eds.)
Proceedings of a Workshop Foundation of Semantic Oriented Data and Web
Mining, held in Conjunction with the Fifth IEEE International Conference on
Data Mining, Houston, Texas, USA, pp. 29–33 (November 27-30, 2005) ISBN
0-9738918-7-4
12. Palancar, J.H., León, R.H., Pagola, J.M., Dı́az, A.H.: A Compressed Vertical Bi-
nary Algorithm for Mining Frequent Patterns. In: the book Data Mining: Foun-
dations and Practice Lin, T.Y., Wasilewska, A., Petry, F., Xie, Y.(eds.) Springer-
Verlag, Accepted for publication (to appear)
13. Orlando, S., Palmerini, P., Perego, R., Silvestri, F.: A Scalable Multi-Strategy Algo-
rithm for Counting Frequent Sets Washington, USA, pp. 19–30. In: Proceedings of
the 5th Workshop on High Performance Data Mining, in conjunction with Second
International SIAM Conference on Data Mining (April 2002)
14. Schuster, A., Wolff, R.: Communication-Efficient Distributed Mining of Association
Rules. Data Mining and Knowledge Discovery, 8(2) (March 2004)
15. Zaki, M.J., Parthasarathy, S., Ogihara, M., Li, W.: New algorithms for fast dis-
covery of association rules. In: Heckerman, D., Mannila, H., Pregibon, D., Uthu-
rusamy, R. (eds.) Proceedings of the Third International Conference on Knowledge
Discovery and Data Mining (KDD-97), p. 283. AAAI Press, Stanford (1997)
16. Zaki, M.J., Parthasarathy, S., Ogihara, M., Li, W.: New Parallel Algorithms for
Fast Discovery of Association Rules. Data Mining and Knowledge Discovery 1(4),
343–373 (1997)
17. Zaki, M.J., Parthasarathy, S., Li, W.: A Localized Algorithm for Parallel Associa-
tion Mining. In: Proceedings of the 9th ACM Symposium on Parallel Algorithms
and Architectures (1997)
18. Zaki, M.J.: Parallel and Distributed Association Mining: A Survey. IEEE Concur-
rency. (October- December 1999)
19. Zaki, M., Parthasarhaty, S., Ogihara, M., Li, W.: Parallel Data Mining for Asso-
ciation Rules on Shared Memory Systems. (February 28, 2001)
20. Zaiane, O.R., El-Hajj, M., Lu, P.: Fast Parallel Association Rule Mining with-
out candidacy generation. Techical Report TR01-12. Department of Computing
Sciences, University of Alberta, Canada (2001)
21. Shintani, T., Kitsuregawa, M.: Parallel Mining Algorithms for Generalized Associ-
ation Rules with Classification Hierarchy. In: Proceedings ACM SIGMOD Interna-
tional Conference on Management of Data, SIGMOD 1998, Seattle, Washington,
USA. (June 2-4, 1998)
22. Sam, E.-H., Karypis, H.G., Kumar, V.: Scalable Parallel Data Mining for Associ-
ation Rules. Department of Computer Science. University of Minnesota (1997)
 Another Random Document on
Scribd Without Any Related Topics
B.C.).But the philo-Laconian Xenophon is still more emphatic in his
condemnation of Sparta. Having described her triumphant and
seemingly unassailable position after the subjugation of Olynthus
and Phlius, he proceeds to say,[166]—“I could produce numerous
other incidents, both in and out of Greece, to prove that the gods
take careful note of impious men and of evil-doers; but the events
which I am now about to relate are quite sufficient. The
Lacedæmonians, who had sworn to leave each city autonomous,
having violated their oaths by seizing the citadel of Thebes, were
punished by the very men whom they had wronged,—though no one
on earth had ever before triumphed over them. And the Theban
faction who had introduced them into the citadel, with the deliberate
purpose that their city should be enslaved to Sparta in order that
they might rule despotically themselves,—were put down by no
more than seven assailants, among the exiles whom they had
banished.”
What must have been the hatred, and sense of abused
ascendency, entertained towards Sparta by neutral or unfriendly
Greeks, when Xenophon, alike conspicuous for his partiality to her
and for his dislike of Thebes, could employ these decisive words in
ushering in the coming phase of Spartan humiliation, representing it
as a well-merited judgment from the gods? The sentence which I
have just translated marks, in the commonplace manner of the
Xenophontic Hellenica, the same moment of pointed contrast and
transition,—past glory suddenly and unexpectedly darkened by
supervening misfortune,—which is foreshadowed in the narrative of
Thucydides by the dialogue between the Athenian envoys and the
Melian[167] council; or in the Œdipus and Antigonê of Sophokles,[168]
by the warnings of the prophet Teiresias.
The government of Thebes had now been for three years (since
the blow struck by Phœbidas) in the hands of Leontiades and his
oligarchical partisans, upheld by the Spartan garrison in the
Kadmeia. Respecting the details of its proceedings we have scarce
any information. We can only (as above remarked) judge of it by the
analogy of the Thirty tyrants at Athens, and of the Lysandrian
Dekarchies, to which it was exactly similar in origin, position, and
interests. That the general spirit of it must have been cruel,
oppressive, and rapacious,—we cannot doubt; though in what
degree we have no means of knowing. The appetites of uncontrolled
rulers, as well as those of a large foreign garrison, would ensure
such a result; besides which, those rulers must have been in
constant fear of risings or conspiracies amidst a body of high-spirited
citizens who saw their city degraded, from being the chief of the
Bœotian federation, into nothing better than a captive dependency
of Sparta. Such fear was aggravated by the vicinity of a numerous
body of Theban exiles, belonging to the opposite or anti-Spartan
party; three or four hundred of whom had fled to Athens at the first
seizure of their leader Ismenias, and had been doubtless joined
subsequently by others. So strongly did the Theban rulers
apprehend mischief from these exiles, that they hired assassins to
take them off by private murder at Athens; and actually succeeded
in thus killing Androkleidas, chief of the band and chief successor of
the deceased Ismenias,—though they missed their blows at the rest.
[169] And we may be sure that they made the prison in Thebes
subservient to multiplied enormities and executions, when we read
not only that one hundred and fifty prisoners were found in it when
the government was put down,[170] but also that in the fervor of that
revolutionary movement, the slain gaoler was an object of such
fierce antipathy, that his corpse was trodden and spit upon by a
crowd of Theban women.[171] In Thebes, as in other Grecian cities,
the women not only took no part in political disputes, but rarely even
showed themselves in public;[172] so that this furious demonstration
of vindictive sentiment must have been generated by the loss or
maltreatment of sons, husbands, and brothers.
The Theban exiles found at Athens not only secure shelter, but
genuine sympathy with their complaints against Lacedæmonian
injustice. The generous countenance which had been shown by the
Thebans, twenty-four years before, to Thrasybulus and the other
Athenian refugees, during the omnipotence of the Thirty, was now
gratefully requited under this reversal of fortune to both cities;[173]
and requited too in defiance of the menaces of Sparta, who
demanded that the exiles should be expelled,—as she had in the
earlier occasion demanded that the Athenian refugees should be
dismissed from Thebes. To protect these Theban exiles, however,
was all that Athens could do. Their restoration was a task beyond
her power,—and seemingly yet more beyond their own. For the
existing government of Thebes was firmly seated, and had the
citizens completely under control. Administered by a small faction,
Archias, Philippus, Hypatês, and Leontiades (among whom the first
two were at this moment polemarchs, though the last was the most
energetic and resolute)—it was at the same time sustained by the
large garrison of fifteen hundred Lacedæmonians and allies,[174]
under Lysanoridas and two other harmosts, in the Kadmeia,—as well
as by the Lacedæmonian posts in the other Bœotian cities around,—
Orchomenus, Thespiæ, Platæa, Tanagra, etc. Though the general
body of Theban sentiment in the city was decidedly adverse to the
government, and though the young men while exercising in the
palæstra (gymnastic exercises being more strenuously prosecuted at
Thebes than anywhere else except at Sparta) kept up by private
communication the ardor of an earnest, but compressed, patriotism,
—yet all manifestation or assemblage was forcibly kept down, and
the commanding posts of the lower town, as well as the citadel,
were held in vigilant occupation by the ruling minority.[175]
For a certain time the Theban exiles at Athens waited in hopes of
some rising at home, or some positive aid from the Athenians. At
length, in the third winter after their flight, they began to despair of
encouragement from either quarter, and resolved to take the
initiative upon themselves. Among them were numbered several
men of the richest and highest families at Thebes, proprietors of
chariots, jockeys, and training establishments, for contending at the
various festivals: Pelopidas, Mellon, Damokleidas, Theopompus,
Pherenikus, and others.[176]
 Of these the most forward in originating aggressive measures,
though almost the youngest, was Pelopidas; whose daring and self-
devotion, in an enterprise which seemed utterly desperate, soon
communicated itself to a handful of his comrades. The exiles,
keeping up constant private correspondence with their friends in
Thebes, felt assured of the sympathy of the citizens generally, if they
could once strike a blow. Yet nothing less would be sufficient than
the destruction of the four rulers, Leontiades and his colleagues,—
nor would any one within the city devote himself to so hopeless a
danger. It was this conspiracy which Pelopidas, Mellon, and five or
ten other exiles (the entire band is differently numbered, by some as
seven, by others, twelve[177]) undertook to execute. Many of their
friends in Thebes came in as auxiliaries to them, who would not
have embarked in the design as primary actors. Of all auxiliaries, the
most effective and indispensable was Phyllidas, the secretary of the
polemarchs; next to him, Charon, an eminent and earnest patriot.
Phyllidas, having been despatched to Athens on official business,
entered into secret conference with the conspirators, concerted with
them the day for their coming to Thebes, and even engaged to
provide for them access to the persons of the polemarchs. Charon
not only promised them concealment in his house, from their first
coming within the gates until the moment of striking their blow
should have arrived,—but also entered his name to share in the
armed attack. Nevertheless, in spite of such partial encouragements,
the plan still appeared desperate to many who wished heartily for its
success. Epaminondas, for example,—who now for the first time
comes before us,—resident at Thebes, and not merely sympathizing
with the political views of Pelopidas, but also bound to him by
intimate friendship,—dissuaded others from the attempt, and
declined participating in it. He announced distinctly that he would
not become an accomplice in civil bloodshed. It appears that there
were men among the exiles whose violence made him fear that they
would not, like Pelopidas, draw the sword exclusively against
Leontiades and his colleagues, but would avail themselves of success
to perpetrate unmeasured violence against other political enemies.
[178]
 The day for the enterprise was determined by Phyllidas the
secretary, who had prepared an evening banquet for Archias and
Philippus, in celebration of the period when they were going out of
office as polemarchs,—and who had promised on that occasion to
bring into their company some women remarkable for beauty, as
well as of the best families in Thebes.[179] In concert with the
general body of Theban exiles at Athens, who held themselves ready
on the borders of Attica, together with some Athenian sympathizers,
to march to Thebes the instant that they should receive intimation,—
and in concert also with two out of the ten Stratêgi of Athens, who
took on themselves privately to countenance the enterprise, without
any public vote,—Pelopidas and Mellon, and their five companions,
[180] crossed Kithæron from Athens to Thebes. It was wet weather,
about December B.C. 379; they were disguised as rustics or hunters,
with no other arms than a concealed dagger; and they got within
the gates of Thebes one by one at nightfall, just when the latest
farming men were coming home from their fields. All of them arrived
safe at the house of Charon, the appointed rendezvous.
It was, however, by mere accident that they had not been turned
back, and the whole scheme frustrated. For a Theban named
Hipposthenidas, friendly to the conspiracy, but faint-hearted, who
had been let into the secret against the will of Phyllidas,—became so
frightened as the moment of execution approached, that he took
upon himself, without the knowledge of the rest, to despatch
Chlidon, a faithful slave of Mellon, ordering him to go forth on
horseback from Thebes, to meet his master on the road, and to
desire that he and his comrades would go back to Attica, since
circumstances had happened to render the project for the moment
impracticable. Chlidon, going home to fetch his bridle, but not
finding it in its usual place, asked his wife where it was. The woman,
at first pretending to look for it, at last confessed that she had lent it
to a neighbor. Chlidon became so irritated with this delay, that he
got into a loud altercation with his wife, who on her part wished him
ill luck with his journey. He at last beat her, until neighbors ran in to
interpose. His departure was thus accidentally frustrated, so that the
intended message of countermand never reached the conspirators
on their way.[181]
In the house of Charon they remained concealed all the ensuing
day, on the evening of which the banquet of Archias and Philippus
was to take place. Phyllidas had laid his plan for introducing them at
that banquet, at the moment when the two polemarchs had become
full of wine, in female attire, as being the women whose visit was
expected. The hour had nearly arrived, and they were preparing to
play their parts, when an unexpected messenger knocked at the
door, summoning Charon instantly into the presence of the
polemarchs. All within were thunderstruck with the summons, which
seemed to imply that the plot had been divulged, perhaps by the
timid Hipposthenidas. It was agreed among them that Charon must
obey at once. Nevertheless, he himself, even in the perilous
uncertainty which beset him, was most of all apprehensive lest the
friends whom he had sheltered should suspect him of treachery
towards themselves and their cause. Before departing, therefore, he
sent for his only son, a youth of fifteen, and of conspicuous promise
in every way. This youth he placed in the hands of Pelopidas, as a
hostage for his own fidelity. But Pelopidas and the rest, vehemently
disclaiming all suspicion, entreated Charon to put his son away, out
of the reach of that danger in which all were now involved. Charon,
however, could not be prevailed on to comply, and left his son
among them to share the fate of the rest. He went into the presence
of Archias and Philippus; whom he found already half-intoxicated,
but informed, by intelligence from Athens, that some plot, they knew
not by whom, was afloat. They had sent for him to question him, as
a known friend of the exiles; but he had little difficulty, aided by the
collusion of Phyllidas, in blinding the vague suspicions of drunken
men, anxious only to resume their conviviality.[182] He was allowed
to retire and rejoin his friends. Nevertheless, soon after his
departure,—so many were the favorable chances which befel these
improvident men,—a fresh message was delivered to Archias the
polemarch, from his namesake Archias the Athenian Hierophant,
giving an exact account of the names and scheme of the
conspirators, which had become known to the philo-Laconian party
at Athens. The messenger who bore this despatch delivered it to
Archias with an intimation, that it related to very serious matters.
“Serious matters for to-morrow,” said the polemarch, as he put the
despatch, unopened and unread, under the pillow of the couch on
which he was reclining.[183]
Returning to their carousal, Archias and Philippus impatiently
called upon Phyllidas to introduce the women according to his
promise. Upon this the secretary retired, and brought the
conspirators, clothed in female attire, into an adjoining chamber;
then going back to the polemarchs, he informed them that the
women would not come in unless all the domestics were first
dismissed. An order was forthwith given that these latter should
depart, while Phyllidas took care that they should be well provided
with wine at the lodging of one among their number. The
polemarchs were thus left only with one or two friends at table, half-
intoxicated as well as themselves; among them Kabeirichus, the
archon of the year, who always throughout his term kept the
consecrated spear of office in actual possession, and had it at that
moment close to his person. Phyllidas now conducted the pretended
women into the banqueting-room; three of them attired as ladies of
distinction, the four others following as female attendants. Their long
veils, and ample folds of clothing, were quite sufficient as disguise,—
even had the guests at table been sober,—until they sat down by the
side of the polemarchs; and the instant of lifting their veils was the
signal for using their daggers. Archias and Philippus were slain at
once and with little resistance; but Kabeirichus with his spear tried to
defend himself, and thus perished with the others, though the
conspirators had not originally intended to take his life.[184]
Having been thus far successful, Phyllidas conducted three of the
conspirators,—Pelopidas, Kephisodôrus, and Damokleidas,—to the
house of Leontiades, into which he obtained admittance by
announcing himself as the bearer of an order from the polemarchs.
Leontiades was reclining after supper, with his wife sitting spinning
wool by his side, when they entered his chamber. Being a brave and
powerful man, he started up, seized his sword, and mortally
wounded Kephisodôrus in the throat; a desperate struggle then
ensued between him and Pelopidas in the narrow doorway, where
there was no room for a third to approach. At length, however,
Pelopidas overthrew and killed him, after which they retired,
enjoining the wife with threats to remain silent, and closing the door
after them with peremptory commands that it should not be again
opened. They then went to the house of Hypatês, whom they slew
while he attempted to escape over the roof.[185]
The four great rulers of the philo-Laconian party in Thebes
having been now put to death, Phyllidas proceeded with the
conspirators to the prison. Here the gaoler, a confidential agent in
the oppressions of the deceased governors, hesitated to admit him;
but was slain by a sudden thrust with his spear, so as to ensure free
admission to all. To liberate the prisoners, probably, for the most
part men of kindred politics with the conspirators,—to furnish them
with arms taken from the battle-spoils hanging up in the neighboring
porticos,—and to range them in battle order near the temple of
Amphion,—were the next proceedings; after which they began to
feel some assurance of safety and triumph.[186] Epaminondas and
Gorgidas, apprised of what had occurred, were the first who
appeared in arms with a few friends to sustain the cause; while
proclamation was everywhere made aloud, through heralds, that the
despots were slain,—that Thebes was free,—and that all Thebans
who valued freedom should muster in arms in the market-place.
There were at that moment in Thebes many trumpeters who had
come to contend for the prize at the approaching festival of the
Herakleia. Hipposthenidas engaged these men to blow their
trumpets in different parts of the city, and thus everywhere to excite
the citizens to arms.[187]
Although during the darkness surprise was the prevalent feeling,
and no one knew what to do,—yet so soon as day dawned, and the
truth became known, there was but one feeling of joy and patriotic
enthusiasm among the majority of the citizens.[188] Both horsemen
and hoplites hastened in arms to the agora. Here for the first time
since the seizure of the Kadmeia by Phœbidas, a formal assembly of
the Theban people was convened, before which Pelopidas and his
fellow-conspirators presented themselves. The priests of the city
crowned them with wreaths, and thanked them in the name of the
local gods; while the assembly hailed them with acclamations of
delight and gratitude, nominating with one voice Pelopidas, Mellon,
and Charon, as the first renewed Bœotarchs.[189] The revival of this
title, which had been dropped since the peace of Antalkidas, was in
itself an event of no mean significance; implying not merely that
Thebes had waked up again into freedom, but that the Bœotian
confederacy also had been, or would be, restored.
Messengers had been forthwith despatched by the conspirators
to Attica to communicate their success; upon which all the remaining
exiles, with the two Athenian generals privy to the plot, and a body
of Athenian volunteers, or corps francs, all of whom were ready on
the borders awaiting the summons,—flocked to Thebes to complete
the work. The Spartan generals, on their side also, sent to Platæa
and Thespiæ for aid. During the whole night, they had been
distracted and alarmed by the disturbance in the city; lights showing
themselves here and there, with trumpets sounding and shouts for
the recent success.[190] Apprised speedily of the slaughter of the
polemarchs, from whom they had been accustomed to receive
orders, they knew not whom to trust or to consult, while they were
doubtless beset by affrighted fugitives of the now defeated party,
who would hurry up the Kadmeia for safety. They reckoned at first
on a diversion in their favor from the forces at Platæa and Thespiæ.
But these forces were not permitted even to approach the city gate;
being vigorously charged, as soon as they came in sight, by the
newly-mustered Theban cavalry, and forced to retreat with loss. The
Lacedæmonians in the citadel were thus not only left without
support, but saw their enemies in the city reinforced by the other
exiles, and by the auxiliary volunteers.[191]
 Meanwhile, Pelopidas and the other new Bœotarchs found
themselves at the head of a body of armed citizens, full of devoted
patriotism and unanimous in hailing the recent revolution. They
availed themselves of this first burst of fervor to prepare for
storming the Kadmeia without delay, knowing the importance of
forestalling all aid from Sparta. And the citizens were already rushing
up to the assault,—proclamation being made of large rewards to
those who should first force their way in,—when the Lacedæmonian
commander sent proposals for a capitulation.[192] Undisturbed
egress from Thebes, with the honors of war, being readily
guaranteed to him by oath, the Kadmeia was then surrendered. As
the Spartans were marching out of the gates, many Thebans of the
defeated party came forth also. But against these latter the
exasperation of the victors was so ungovernable, that several of the
most odious were seized as they passed, and put to death; in some
cases, even their children along with them. And more of them would
have been thus despatched, had not the Athenian auxiliaries, with
generous anxiety, exerted every effort to get them out of sight and
put them into safety.[193] We are not told,—nor is it certain,—that
these Thebans were protected under the capitulation. Even had they
been so, however, the wrathful impulse might still have prevailed
against them. Of the three harmosts who thus evacuated the
Kadmeia without a blow, two were put to death, the third was
heavily fined and banished, by the authorities at Sparta.[194] We do
not know what the fortifications of the Kadmeia were, nor how far it
was provisioned. But we can hardly wonder that these officers were
considered to have dishonored the Lacedæmonian arms, by making
no attempt to defend it; when we recollect that hardly more than
four or five days would be required to procure adequate relief from
home,—and that forty-three years afterwards, the Macedonian
garrison in the same place maintained itself against the Thebans in
the city for more than fourteen days, until the return of Alexander
from Illyria.[195] The first messenger who brought news to Sparta of
the conspiracy and revolution at Thebes, appears to have
communicated at the same time that the garrison had evacuated the
Kadmeia and was in full retreat, with a train of Theban exiles from
the defeated party.[196]
This revolution at Thebes came like an electric shock upon the
Grecian world. With a modern reader, the assassination of the four
leaders, in their houses and at the banquet, raises a sentiment of
repugnance which withdraws his attention from the other features of
this memorable deed. Now an ancient Greek not only had no such
repugnance, but sympathized with the complete revenge for the
seizure of the Kadmeia and the death of Ismenias; while he admired,
besides, the extraordinary personal daring of Pelopidas and Mellon,
—the skilful forecast of the plot,—and the sudden overthrow, by a
force so contemptibly small, of a government which the day before
seemed unassailable.[197] It deserves note that we here see the
richest men in Thebes undertaking a risk, single-handed and with
their own persons, which must have appeared on a reasonable
estimate little less than desperate. From the Homeric Odysseus and
Achilles down to the end of free Hellenism, the rich Greek strips in
the Palæstra,[198] and exposes his person in the ranks as a soldier
like the poorest citizens; being generally superior to them in strength
and bodily efficiency.
As the revolution in Thebes acted forcibly on the Grecian mind
from the manner in which it was accomplished, so by its positive
effects it altered forthwith the balance of power in Greece. The
empire of Sparta, far from being undisputed and nearly universal
over Greece, is from henceforward only maintained by more or less
effort, until at length it is completely overthrown.[199]
The exiles from Thebes, arriving at Sparta, inflamed both the
ephors, and the miso-Theban Agesilaus, to the highest pitch. Though
it was then the depth of winter,[200] an expedition was decreed
forthwith against Thebes, and the allied contingents were
summoned. Agesilaus declined to take the command of it, on the
ground that he was above sixty years of age, and therefore no
longer liable to compulsory foreign service. But this (says
Xenophon[201]) was not his real reason. He was afraid that his
enemies at Sparta would say,—“Here is Agesilaus again putting us to
expense, in order that he may uphold despots in other cities,”—as he
had just done, and had been reproached with doing, at Phlius; a
second proof that the reproaches against Sparta (which I have cited
a few pages above from Lysias and Isokrates) of allying herself with
Greek despots as well as with foreigners to put down Grecian
freedom, found an echo even in Sparta herself. Accordingly
Kleombrotus, the other king of Sparta, took the command. He had
recently succeeded his brother Agesipolis, and had never
commanded before.
Kleombrotus conducted his army along the Isthmus of Corinth
through Megara to Platæa, cutting to pieces an outpost of Thebans,
composed chiefly of the prisoners set free by the recent revolution,
who had been placed for the defence of the intervening mountain-
pass. From Platæa he went forward to Thespiæ, and from thence to
Kynoskephalæ in the Theban territory, where he lay encamped for
sixteen days; after which he retreated to Thespiæ. It appears that
he did nothing, and that his inaction was the subject of much
wonder in his army, who are said to have even doubted whether he
was really and earnestly hostile to Thebes. Perhaps the exiles, with
customary exaggeration, may have led him to hope that they could
provoke a rising in Thebes, if he would only come near. At any rate
the bad weather must have been a serious impediment to action;
since in his march back to Peloponnesus through Kreusis and
Ægosthenæ the wind blew a hurricane, so that his soldiers could not
proceed without leaving their shields and coming back afterwards to
fetch them. Kleombrotus did not quit Bœotia, however, without
leaving Sphodrias as harmost at Thespiæ, with one third of the
entire army, and with a considerable sum of money to employ in
hiring mercenaries and acting vigorously against the Thebans.[202]
The army of Kleombrotus, in its march from Megara to Platæa,
had passed by the skirts of Attica; causing so much alarm to the
Athenians, that they placed Chabrias with a body of peltasts, to
guard their frontier and the neighboring road through Eleutheræ into
Bœotia. This was the first time that a Lacedæmonian army had
touched Attica (now no longer guarded by the lines of Corinth, as in
the war between 394 and 388 B.C.) since the retirement of king
Pausanias in 404 B.C.; furnishing a proof of the exposure of the
country, such as to revive in the Athenian mind all the terrible
recollections of Dekeleia and the Peloponnesian war. It was during
the first prevalence of this alarm,—and seemingly while Kleombrotus
was still with his army at Thespiæ or Kynoskephalæ, close on the
Athenian frontier,—that three Lacedæmonian envoys, Etymoklês and
two others, arrived at Athens to demand satisfaction for the part
taken by the two Athenian generals and the Athenian volunteers, in
concerting and aiding the enterprise of Pelopidas and his comrades.
So overpowering was the anxiety in the public mind to avoid giving
offence to Sparta, that these two generals were both of them
accused before the dikastery. The first of them was condemned and
executed; the second, profiting by this warning (since, pursuant to
the psephism of Kannônus,[203] the two would be put on trial
separately), escaped, and a sentence of banishment was passed
against him.[204] These two generals had been unquestionably guilty
of a grave abuse of their official functions. They had brought the
state into public hazard, not merely without consulting the senate or
assembly, but even without taking the sense of their own board of
Ten. Nevertheless the severity of the sentence pronounced indicates
the alarm, as well as the displeasure, of the general body of
Athenians; while it served as a disclaimer in fact, if not in form, of all
political connection with Thebes.[205]
Even before the Lacedæmonian envoys had quitted Athens,
however, an incident, alike sudden and memorable, completely
altered the Athenian temper. The Lacedæmonian harmost Sphodrias
(whom Kleombrotus had left at Thespiæ to prosecute the war
against Thebes), being informed that Peiræus on its land side was
without gates or night watch,—since there was no suspicion of
attack,—conceived the idea of surprising it by a night-march from
Thespiæ, and thus of mastering at one stroke the commerce, the
wealth, and the naval resources of Athens. Putting his troops under
march one evening after an early supper, he calculated on reaching
the Peiræus the next morning before daylight. But his reckoning
proved erroneous. Morning overtook him when he had advanced no
farther than the Thriasian plain near Eleusis; from whence, as it was
useless to proceed farther, he turned back and retreated to Thespiæ;
not, however, without committing various acts of plunder against the
neighboring Athenian residents.
This plan against Peiræus appears to have been not ill conceived.
Had Sphodrias been a man competent to organize and execute
movements as rapid as those of Brasidas, there is no reason why it
might not have succeeded; in which case the whole face of the war
would have been changed, since the Lacedæmonians, if once
masters of Peiræus, both could and would have maintained the
place. But it was one of those injustices, which no one ever
commends until it has been successfully consummated,—“consilium
quod non potest laudari nisi peractum.[206]” As it failed, it has been
considered, by critics as well as by contemporaries, not merely as a
crime but as a fault, and its author Sphodrias as a brave man, but
singularly weak and hot-headed.[207] Without admitting the full
extent of this censure, we may see that his present aggression grew
out of an untoward emulation of the glory which Phœbidas, in spite
of the simulated or transient displeasure of his countrymen, had
acquired by seizing the Kadmeia. That Sphodrias received private
instructions from Kleombrotus (as Diodorus states) is not sufficiently
proved; while the suspicion, intimated by Xenophon as being abroad,
that he was wrought upon by secret emissaries and bribes from his
enemies the Thebans, for the purpose of plunging Athens into war
with Sparta, is altogether improbable;[208] and seems merely an
hypothesis suggested by the consequences of the act,—which were
such, that if his enemies had bribed him, he could not have served
them better.
The presence of Sphodrias and his army in the Thriasian plain
was communicated shortly after daybreak at Athens, where it
excited no less terror than surprise. Every man instantly put himself
under arms for defence; but news soon arrived that the invader had
retired. When thus reassured, the Athenians passed from fear to
indignation. The Lacedæmonian envoys, who were lodging at the
house of Kallias the proxenus of Sparta, were immediately put under
arrest and interrogated. But all three affirmed that they were not
less astonished, and not less exasperated, by the march of
Sphodrias, than the Athenians themselves; adding, by way of
confirmation, that had they been really privy to any design of seizing
the Peiræus, they would have taken care not to let themselves be
found in the city, and in their ordinary lodging at the house of the
proxenus, where of course their persons would be at once seized.
They concluded by assuring the Athenians, that Sphodrias would not
only be indignantly disavowed, but punished capitally, at Sparta. And
their reply was deemed so satisfactory, that they were allowed to
depart; while an Athenian embassy was sent to Sparta, to demand
the punishment of the offending general.[209]
The Ephors immediately summoned Sphodrias home to Sparta,
to take his trial on a capital charge. So much did he himself despair
of his case, that he durst not make his appearance; while the
general impression was, both at Sparta and elsewhere, that he
would certainly be condemned. Nevertheless, though thus absent
and undefended, he was acquitted, purely through private favor and
esteem for his general character. He was of the party of
Kleombrotus, so that all the friends of that prince espoused his
cause, as a matter of course. But as he was of the party opposed to
Agesilaus, his friends dreaded that the latter would declare against
him, and bring about his condemnation. Nothing saved Sphodrias
except the peculiar intimacy between his son Kleonymus and
Archidamus son of Agesilaus. The mournful importunity of
Archidamus induced Agesilaus, when this important cause was
brought before the Senate of Sparta, to put aside his judicial
conviction, and give his vote in the following manner: “To be sure,
Sphodrias is guilty; upon that there cannot be two opinions.
Nevertheless, we cannot put to death a man like him, who, as boy,
youth, and man, has stood unblemished in all Spartan honor. Sparta
cannot part with soldiers like Sphodrias.[210]” The friends of
Agesilaus, following this opinion and coinciding with those of
Kleombrotus, ensured a favorable verdict. And it is remarkable, that
Etymoklês himself, who as envoy at Athens had announced as a
certainty that Sphodrias would be put to death,—as senator and
friend of Agesilaus voted for his acquittal.[211]
This remarkable incident (which comes to us from a witness not
merely philo-Laconian, but also personally intimate with Agesilaus)
shows how powerfully the course of justice at Sparta was overruled
by private sympathy and interests,—especially, those of the two
kings. It especially illustrates what has been stated in a former
chapter respecting the oppressions exercised by the Spartan
harmosts and the dekadarchies, for which no redress was attainable
at Sparta. Here was a case where not only the guilt of Sphodrias
stood confessed, but in which also his acquittal was sure to be
followed by a war with Athens. If, under such circumstances, the
Athenian demand for redress was overruled by the favor of the two
kings, what chance was there of any justice to the complaint of a
dependent city, or an injured individual, against the harmost? The
contrast between Spartan and Athenian proceeding is also
instructive. Only a few days before, the Athenians condemned, at
the instance of Sparta, their two generals who had without authority
lent aid to the Theban exiles. In so doing, the Athenian dikastery
enforced the law against clear official misconduct,—and that, too, in
a case where their sympathies went along with the act, though their
fear of a war with Sparta was stronger. But the most important
circumstance to note is, that at Athens there is neither private
influence, nor kingly influence, capable of overruling the sincere
judicial conscience of a numerous and independent dikastery.
The result of the acquittal of Sphodrias must have been well
known beforehand to all parties at Sparta. Even by the general voice
of Greece, the sentence was denounced as iniquitous.[212] But the
Athenians, who had so recently given strenuous effect to the
remonstrances of Sparta against their own generals, were stung by
it to the quick; and only the more stung, in consequence of the
extraordinary compliments to Sphodrias on which the acquittal was
made to turn. They immediately contracted hearty alliance with
Thebes, and made vigorous preparations for war against Sparta both
by land and sea. After completing the fortifications of Peiræus, so as
to place it beyond the reach of any future attempt, they applied
themselves to the building of new ships of war, and to the extension
of their naval ascendency, at the expense of Sparta.[213]
From this moment, a new combination began in Grecian politics.
The Athenians thought the moment favorable to attempt the
construction of a new confederacy, analogous to the Confederacy of
Delos, formed a century before; the basis on which had been reared
the formidable Athenian empire, lost at the close of the
Peloponnesian war. Towards such construction there was so far a
tendency, that Athens had already a small body of maritime allies;
while rhetors like Isokrates (in his Panegyrical Discourse, published
two years before) had been familiarizing the public mind with larger
ideas. But the enterprise was now pressed with the determination
and vehemence of men smarting under recent insult. The Athenians
had good ground to build upon; since, while the discontent against
the ascendency of Sparta was widely spread, the late revolution in
Thebes had done much to lessen that sentiment of fear upon which
such ascendency chiefly rested. To Thebes, the junction with Athens
was preëminently welcome, and her leaders gladly enrolled their city
as a constituent member of the new confederacy.[214] They
cheerfully acknowledged the presidency of Athens,—reserving,
however, tacitly or expressly, their own rights as presidents of the
Bœotian federation, as soon as that could be reconstituted; which
reconstitution was at this moment desirable even for Athens, seeing
that the Bœotian towns were now dependent allies of Sparta under
harmosts and oligarchies.
The Athenians next sent envoys round to the principal islands
and maritime cities in the Ægean, inviting all of them to an alliance
on equal and honorable terms. The principles were in the main the
same as those upon which the confederacy of Delos had been
formed against the Persians, almost a century before. It was
proposed that a congress of deputies should meet at Athens, one
from each city, small as well as great, each with one vote; that
Athens should be president, yet each individual city autonomous;
that a common fund should be raised, with a common naval force,
through assessment imposed by this congress upon each, and
applied as the same authority might prescribe; the general purpose
being defined to be, maintenance of freedom and security from
foreign aggression, to each confederate, by the common force of all.
Care was taken to banish as much as possible those associations of
tribute and subjection which rendered the recollection of the former
Athenian empire unpopular.[215] And as there were many Athenian
citizens, who, during those times of supremacy, had been planted
out as kleruchs or out-settlers in various dependencies, but had been
deprived of their properties at the close of the war,—it was thought
necessary to pass a formal decree,[216] renouncing and barring all
revival of these suspended rights. It was farther decreed that
henceforward no Athenian should on any pretence hold property,
either in house or land, in the territory of any one of the
confederates; neither by purchase, nor as security for money lent,
nor by any other mode of acquisition. Any Athenian infringing this
law, was rendered liable to be informed against before the synod;
who, on proof of the fact, were to deprive him of the property,—half
of it going to the informer, half to the general purposes of the
confederacy.
Such were the liberal principles of confederacy now proposed by
Athens,—who, as a candidate for power, was straightforward and
just, like the Herodotean Deiokês,[217]—and formally ratified, as well
by the Athenians as by the general voice of the confederate deputies
assembled within their walls. The formal decree and compact of
alliance was inscribed on a stone column and placed by the side of
the statue of Zeus Eleutherius or the Liberator; a symbol, of
enfranchisement from Sparta accomplished, as well as of freedom to
be maintained against Persia and other enemies.[218] Periodical
meetings of the confederate deputies were provided to be held (how
often, we do not know) at Athens, and the synod was recognized as
competent judge of all persons, even Athenian citizens, charged with
treason against the confederacy. To give fuller security to the
confederates generally, it was provided in the original compact, that
if any Athenian citizen should either speak, or put any question to
the vote, in the Athenian assembly, contrary to the tenor of that
document,—he should be tried before the synod for treason; and
that, if found guilty, he might be condemned by them to the severest
punishment.
Three Athenian leaders stood prominent as commissioners in the
first organization of the confederacy, and in the dealings with those
numerous cities whose junction was to be won by amicable
inducement,—Chabrias, Timotheus son of Konon, and Kallistratus.
[219]

The first of the three is already known to the reader. He and

Iphikrates were the most distinguished warriors whom Athens
numbered among her citizens. But not having been engaged in any
war, since the peace of Antalkidas in 387 B.C., she had had no need
of their services; hence both of them had been absent from the city
during much of the last nine years, and Iphikrates seems still to
have been absent. At the time when that peace was concluded,
Iphikrates was serving in the Hellespont and Thrace, Chabrias with
Evagoras in Cyprus; each having been sent thither by Athens at the
head of a body of mercenary peltasts. Instead of dismissing their
troops, and returning to Athens as peaceful citizens, it was not less
agreeable to the military tastes of these generals, than conducive to
their importance and their profit, to keep together their bands, and
to take foreign service. Accordingly, Chabrias had continued in
service first in Cyprus, next with the native Egyptian king Akoris. The
Persians, against whom he served, found his hostility so
inconvenient, that Pharnabazus demanded of the Athenians to recall
him, on pain of the Great King’s displeasure; and requested at the
same time that Iphikrates might be sent to aid the Persian satraps in
organizing a great expedition against Egypt. The Athenians, to
whom the goodwill of Persia was now of peculiar importance,
complied on both points; recalled Chabrias, who thus became
disposable for the Athenian service,[220] and despatched Iphikrates
to take command along with the Persians.
Iphikrates, since the peace of Antalkidas, had employed his
peltasts in the service of the kings of Thrace: first of Seuthes, near
the shores of the Propontis, whom he aided in the recovery of
certain lost dominions,—next of Kotys, whose favor he acquired, and
whose daughter he presently married.[221] Not only did he enjoy
great scope for warlike operations and plunder, among the “butter-
eating Thracians,”[222]—but he also acquired, as dowry, a large stock
of such produce as Thracian princes had at their disposal, together
with a boon even more important,—a seaport village not far from
the mouth of the Hebrus, called Drys, where he established a
fortified post, and got together a Grecian colony dependent on
himself.[223] Miltiades, Alkibiades, and other eminent Athenians had
done the same thing before him; though Xenophon had refused a
similar proposition when made to him by the earlier Seuthes.[224]
Iphikrates thus became a great man in Thrace, yet by no means
abandoning his connection with Athens, but making his position in
each subservient to his importance in the other. While he was in a
situation to favor the projects of Athenian citizens for mercantile and
territorial acquisitions in the Chersonese and other parts of Thrace,—
he could also lend the aid of Athenian naval and military art, not
merely to princes in Thrace, but to others even beyond those limits,
—since we learn that Amyntas king of Macedonia became so
attached or indebted to him as to adopt him for his son.[225] When
sent by the Athenians to Persia, at the request of Pharnabazus
(about 378 B.C. apparently), Iphikrates had fair ground for
anticipating that a career yet more lucrative was opening before him.
[226]

Iphikrates being thus abroad, the Athenians joined with Chabrias,

in the mission and measures for organizing their new confederacy,
two other colleagues, of whom we now hear for the first time—
Timotheus son of Konon, and Kallistratus the most celebrated orator
of his time.[227] The abilities of Kallistratus were not military at all;
while Timotheus and Chabrias were men of distinguished military
merit. But in acquiring new allies and attracting deputies to her
proposed congress, Athens stood in need of persuasive appeal,
conciliatory dealing, and substantial fairness in all her propositions,
not less than of generalship. We are told that Timotheus, doubtless
as son of the liberator Konon, from the recollections of the battle of
Knidus—was especially successful in procuring new adhesions; and
probably Kallistratus,[228] going round with him to the different
islands, contributed by his eloquence not a little to the same result.
On their invitation, many cities entered as confederates.[229] At this
time (as in the earlier confederacy of Delos) all who joined must
have been unconstrained members. And we may understand the
motives of their junction, when we read the picture drawn by
Isokrates (in 380 B.C.) of the tyranny of the Persians on the Asiatic
mainland, threatening, to absorb the neighboring islands. Not only
was there now a new basis of imposing force, presented by Athens
and Thebes in union—but there was also a wide-spread hatred of
imperial Sparta, aggravated since her perversion of the pretended
boon of autonomy, promised by the peace of Antalkidas; and the
conjunction of these sentiments caused the Athenian mission of
invitation to be extremely successful. All the cities in Eubœa (except
Histiæa, at the north of the island)—as well as Chios, Mitylênê,
Byzantium, and Rhodes—the three former of whom had continued
favorably inclined to Athens ever since the peace of Antalkidas,[230]
—all entered into the confederacy. An Athenian fleet under Chabrias,
sailing among the Cyclades and the other islands of the Ægean,
aided in the expulsion of the Lacedæmonian harmosts,[231] together
with their devoted local oligarchies, wherever they still subsisted;
and all the cities thus liberated became equal members of the newly-
constituted congress at Athens. After a certain interval, there came
to be not less than seventy cities, many of them separately powerful,
which sent deputies to it;[232] an aggregate sufficient to intimidate
Sparta, and even to flatter Athens with the hope of restoration to
something like her former lustre.
The first votes both of Athens herself, and of the newly-
assembled congress, threatened war upon the largest scale. A
resolution was passed to equip twenty thousand hoplites, five
hundred horsemen, and two hundred triremes.[233] Probably the
insular and Ionic deputies promised each a certain contribution of
money, but nothing beyond. We do not, however, know how much,
—nor how far the engagements, large or small, were realized,—nor
whether Athens was authorized to enforce execution against
defaulters,—or was in circumstances to act upon such authority, if
granted to her by the congress. It was in this way (as the reader will
recollect from my fifth volume) that Athens had first rendered herself
unpopular in the confederacy of Delos,—by enforcing the resolutions
of the confederate synod against evasive or seceding members. It
was in this way that what was at first a voluntary association had
ultimately slid into an empire by constraint. Under the new
circumstances of 378 B.C., we may presume that the confederates,
though ardent and full of promises on first assembling at Athens,
were even at the outset not exact, and became afterwards still less
exact, in performance; yet that Athens was forced to be reserved in
claiming, or in exercising, the right of enforcement. To obtain a vote
of contribution by the majority of deputies present, was only the first
step in the process; to obtain punctual payment, when the Athenian
fleet was sent round for the purpose of collecting,—yet without
incurring dangerous unpopularity,—was the second step, but by far
the most doubtful and difficult.
It must, however, be borne in mind that at this moment, when
the confederacy was first formed, both Athens and the other cities
came together from a spontaneous impulse of hearty mutuality and
coöperation. A few years afterwards, we shall find this changed;
Athens selfish, and the confederates reluctant.[234] Inflamed, as well
by their position of renovated headship, as by fresh animosity
against Sparta, the Athenians made important efforts of their own,
both financial and military. Equipping a fleet, which for the time was
superior in the Ægean, they ravaged the hostile territory of Histiæa
in Eubœa, and annexed to their confederacy the islands of
Peparêthus and Skiathus. They imposed upon themselves also a
direct property-tax; to what amount, however, we do not know.
It was on the occasion of this tax that they introduced a great
change in the financial arrangements and constitution of the city; a
change conferring note upon the archonship of Nausinikus, (B.C. 378-
377). The great body of substantial Athenian citizens as well as
metics were now classified anew for purposes of taxation. It will be
remembered that even from the time of Solon[235] the citizens of
Athens had been distributed into four classes,—Pentakosiomedimni,
Hippeis, Zeugitæ, Thêtes,—distinguished from each other by the
amount of their respective properties. Of these Solonian classes, the
fourth, or poorest, paid no direct taxes; while the three former were
taxed according to assessments representing a certain proportion of
their actual property. The taxable property of the richest (or
Pentakosiomedimni, including all at or above the minimum income of
five hundred medimni of corn per annum) was entered in the tax-
book at a sum equal to twelve times their income; that of the
Hippeis (comprising all who possessed between three hundred and
five hundred medimni of annual income) at ten times their income;
that of the Zeugitæ (or possessors of an annual income between
two hundred and three hundred medimni) at five times their income.
A medimnus of corn was counted as equivalent to a drachma; which
permitted the application of this same class-system to movable
property as well as to land. So that, when an actual property-tax (or
eisphora) was imposed, it operated as an equal or proportional tax,
so far as regarded all the members of the same class; but as a
graduated or progressive tax, upon all the members of the richer
class as compared with those of the poorer.
The three Solonian property-classes above named appear to have
lasted, though probably not without modifications, down to the close
of the Peloponnesian war; and to have been in great part preserved,
after the renovation of the democracy in B.C. 403, during the
archonship of Eukleides.[236] Though eligibility to the great offices of
state had before that time ceased to be dependent on pecuniary
qualification, it was still necessary to possess some means of
distinguishing the wealthier citizens, not merely in case of direct
taxation being imposed, but also because the liability to serve in
liturgies or burdensome offices was consequent on a man’s
enrolment as possessor of more than a given minimum of property.
It seems, therefore, that the Solonian census, in its main principles
of classification and graduation, was retained. Each man’s property
being valued, he was ranged in one of three or more classes
according to its amount. For each of the classes, a fixed proportion
of taxable capital to each man’s property was assumed, and each
was entered in the schedule, not for his whole property, but for the
sum of taxable capital corresponding to his property, according to
the proportion assumed. In the first or richest class, the taxable
capital bore a greater ratio to the actual property than in the less
rich; in the second, a greater ratio than in the third. The sum of all
these items of taxable capital, in all the different classes, set
opposite to each man’s name in the schedule, constituted the
aggregate census of Attica; upon which all direct property-tax was
imposed, in equal proportion upon every man.
Respecting the previous modifications in the register of taxable
property, or the particulars of its distribution into classes, which had
been introduced in 403 B.C. at the archonship of Eukleides, we have
no information. Nor can we make out how large or how numerous
were the assessments of direct property-tax, imposed at Athens
between that archonship and the archonship of Nausinikus in 378
B.C. But at this latter epoch the register was again considerably
modified, at the moment when Athens was bracing herself up for
increased exertions. A new valuation was made of the property of
every man possessing property to the amount of twenty-five minæ
(or twenty-five hundred drachmæ) and upwards. Proceeding upon
this valuation, every one was entered in the schedule for a sum of
taxable capital equal to a given fraction of what he possessed. But
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookball.com