International Journal of Computer Systems (ISSN: 2394-1065), Volume 03 Issue 05, May, 2016

Available at http://www.ijcsonline.com/

Classification of Task Partitioning and Load Balancing Strategies in Distributed

Parallel Computing Systems
Maninderjit Singh KhannaA, Dr. Jitendra SheetlaniB

Computer Application Dept., Chitkara University, Rajpura, Punjab (India)

Computer Application Dept., Satya Sai University Tech. and Medical Science, Sehore, Bhopal (MP) India

Abstract
Distribution of the tasks amongst the various computing nodes is itself an intellectually challenging problem in the high
performance distributed computing systems. To choose the appropriate strategy for the required system is difficult
without the meaningful comparison of the existing task partitioning and load balancing strategies. The effectiveness of
the strategy depend on the number of factors-efficiency, interconnection topology, communication mode, program
structure, throughput and computing capabilities of the structure. A number of task partitioning and load balancing
strategies have been proposed, each of which perform remarkable results under different circumstances. The main goal
of the paper is to unravel the mystery of strategies and to classify when and where each strategy is appropriate. In this
paper, taxonomy of task partitioning and load balancing is presented in an attempt to provide a common terminology
and classification mechanism.
Keywords: Task Scheduling, Dynamic, Preemptive, Non-Preemptive, Parallel Computing etc.

I.

INTRODUCTION

Scheduling is a function that assign job to the different

processors [6]. Its a two-step process, namely processor
allocation and assignment. Jobs are autonomous program
that execute in its domain. Resource allocation is done by
the scheduler over two dimensions, time and space, and at
two level jobs and threads. In running state, job constitutes
threads to reduce the overhead. If software package is used
to execute the parallel jobs instead of threads, it increased
load of the parallel computing system. Interacting entities
are called process if they are independent under the
limitation of the operating system. Threads and
communication between them may be static and dynamic
[10].For example, in MIMD architecture [1], number of
threads, and the communication pattern between the
threads can change dynamically during the execution in the
parallel computing systems. If multiple executing entities
are parts of the same application then we treat entities as a
threads and application as a job. Parallel computing is used
to solve the large problems in the scientific and systematic
manner. Effective task partitioning and load balancing
strategies of large task is crucial to achieve high
performance in parallel and distributed system. The
increasing demand of the high performance computing
systems amongst the various field of the science is the key
feature of interest in the parallel computing. The selection
of appropriate strategy for the particular system is the
deciding factor for the successful execution of the tasks. A
parallel job is the collection of the tasks having some
precedence relationship. A task can be identified as a
executable fragment that must be sequentially executed
without partial parallel execution [2].All parallel job cannot
be fully parallelized. In a homogeneous architecture the
serial fraction of the computation has to be executed at the
speed of any of the identical processor, limiting the speed
that can be obtain due to parallel execution of the jobs. The

sequential bottleneck can be greatly reduced by executing

the tasks on the heterogeneous parallel computers by
running the critical tasks on a faster processor. The
efficiency analysis of the heterogeneity presented by
Polychronopoulos and Andrews [4] and Almeida and
Menasce [5]. Processor allocation in the efficient manner
deals with the determination of the number of processor
allocated to a job; the assignment phase means the
distribution of the jobs task to the allocated processor [7].
Time complexity analysis tool for the task partitioning
developed by Pugh and Nirkhe [8] which accurately
estimates the execution time of general real time program
employing on high level structures. Towsley and Nelson
[9] present an analytical model for partitioning the
independent tasks on the different processors. Due to the
large variety of tools, computing architectures and different
requirement there is no ideal task partitioning strategy.
Impact of heterogeneity in loosely coupled systems
analyzed by Freund in different processing paradigms
(MIMD, SIMD, and SISD etc.)

II.

DETERMINISTIC TASK (STATIC)

PARTITIONING STRATEGIES

In deterministic task partitioning the characteristics of

an application such as communication cost, execution time,
synchronization, and data dependency are known in
advance [11].It assign the tasks to the different processors
before execution, and allocation remain same during the
execution of the process. Dynamic scheduling performs
scheduling at runtime, and allocation of the process may
change during the execution of tasks. Static scheduling
cannot support load balancing and fault tolerance while
dynamic
scheduling
support
these
parameters.
Deterministic scheduling on Network of Workstation
(NOW) is NP-complete in strong sense while
nondeterministic is not strong NP-complete. Purely static

371 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 05, May, 2016

Maninderjit Singh Khanna et al

Classification of Task Partitioning and Load Balancing Strategies in Distributed Parallel Computing Systems

task partitioning with Open CL, determines the best

partitioning across the processors in a system and divides
the tasks into as many chunks as there are processor with
each processor receiving the appropriate chunks [13].

nodes. An important aspect of this node is that, in the case

of load migration each node knows exactly where to send
its extra work load. The communication is take place only
between the neighbor nodes.

PY (Papadimitriou and Yannakakis) [34] scheduling

approximate the absolute achievable lower bound of the
start time of a node by using attribute e-value. From the
first node to exit node e-value is computed recursively to
assign the priority to the node. After calculating the e-value
of the node, each node is inserted into a cluster in such a
way that ancestors have data arrival times larger than the evalue of the node.

Evolutionary Task Scheduling [26] upon the previous

scheduling stages, in static and dynamic environment. The
use of heuristic in the initialization phase and specific
mutation operator are two parameters which provide the
beneficial result for the effective scheduling in the static
environment. In the consistent environment, moving the
tasks from high level processors to the processor of smaller
make span by using rebalancing operator provide the
better result. But in the case of inconsistent environment
less greedy operator provide the good scheduling. The
information collected from the previous state is used, to
select the environment by the scheduler.

LCTD (Linear Clustering with Task Duplication) [35]

scheduling identify the edges among clusters that
determine the completion time after the linear clustering of
DAG. After that it tries to duplicate the parents
corresponding to these edges to reduce the start times of
some nodes in the cluster.
EZ (Edge Zeroing) Scheduling [33] select computing
environment by merging edge weights. This scheduling
strategy finds the edge with the largest weight in every
step. If merging does not increase the completion time, the
two cluster incident by the edges will be merged (zeroing
the largest weight).The ordering of the node is define in the
resulting cluster based on the static b-level of the nodes.
DAG edges are sorted in a descending order of edge
weight.
MCP (Modified Critical Path) [31] scheduling decide
the priority of the nodes on the basis of ALAP. It compute
the ALAP of the existing node, and construct a list of node
in ascending order of ALAP times. MCP looks for the
processor idle time slot for a given node. Select the first
node from the node list and insert it into the processor of
earliest execution time. It cannot guarantee an optimal
schedule for the fork and join structure.
ETF (Earliest Time First) scheduling [32] technique,
select the nods of the smallest start time from the node of
the earliest start time in the ready queue at each step.
Earliest start time is computed by examine the start time of
the node on all processor exhaustively. When two node
have the same EST then ETF breaks the tie by selecting the
node with the higher static priority. The priorities are
assigning to the node by computing the static b-level. This
strategy is not guaranteed an optimal schedule for the fork
and join structure.

III.

NON DETERMINISTIC TASK (DYNAMIC)

PARTITIONING STRATEGIES

Position Scan Task Scheduling (PSTS) is pure dynamic

load balancing, highly parallel and efficient, and can be
used in centralized and decentralized manner [14]. It is
based on divide and conquers principle in which hypergrids of dimension k are divided into grids of dimension k1, until the dimension is 1.Position Scan Load Balancing
(PSLB) technique is two phase technique. In first phase the
system is modeled as a hypercube only at one time. In the
second phase load balancing is performed by obtaining
necessary information for each system node. On the basis
of the prior information PSLB strategy calculate its future
load, and load is migrated according to the capability of the

DPS(Dynamic Priority Scheduling) [27] assign the

priorities to the tasks based on the difference of bottom
level (b-level) and top(t-level), to schedule the minimum
schedule length in Directed Acyclic Graph(DAGs)
[3].Compile time scheduling is used in the heterogeneous
environment .Dynamic process assigning select more
important tasks before less important ones. The mapping
and scheduling strategies are depend upon the processor
scheduler, network architecture and the DAG
structure[28].The t-level is the length of longest path
(execution cost+ communication cost) between target task
and entry task, while b-level is the largest path between
target task and exit task. In this way, t-level determine the
earliest start time, b-level of a task is bounded by the
critical path of the DAG [12].
DLS [29] used a GDL (generalized dynamic level) to
determine dynamic priorities, for all tasks in the ready
queue, over all processors at every scheduling step. GDL
give the priority to the processor according to their speed,
and many other factors are taken into account. Top level
and bottom level are loss their meaning in the case of the
heterogeneous architecture due to the different execution
cost on each processor. It consider the various factors,
median of the execution cost over all processors, average
execution cost and maximum or minimum cost for the
computation of the schedule levels.

IV.

PREEMPTIVE TASK PARTITIONING

STRATEGIES

Preemptive Deterministic Scheduling (PDS) ensure

replica behavior while preserving concurrency. Replication
is achieved by the threads and there is no communication
between the threads. Performance improvement is achieved
due to multithreading by exploiting concurrency in thread
execution. Multithreaded replica shows non deterministic
behavior. Only one physical thread is schedule at given
time, so it shows poor scalability and performance. PDS
schedule multiple threads at same time so its throughput
five times from Preemptive Nondeterministic Scheduler
(NPDS) [11].PDS remove the need for inter replica
communication yet preserves a large degree of replica
concurrency. Optimal preemptive schedule subject to
release date are taken to minimize the execution time on
the homogeneous platforms. This strategy performs the
two step, in first step, minimize maximum completion time

372 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 05, May, 2016

Maninderjit Singh Khanna et al

Classification of Task Partitioning and Load Balancing Strategies in Distributed Parallel Computing Systems

on the desired number of machines. In second step,

maximum lateness is minimizes with respect to due dates
for the jobs [12] on the arbitrary number of machines.
Fast Preemptive Scheduling (FPS) [17] strategy
simulates preemptive task execution at a very low overhead
and requires small runtime support in heterogeneous and
homogeneous parallel computing environment. Preemptive
Task Scheduling (PTS) [18], a list heuristic scheduling
strategy is used homogeneous distributed memory systems.
RM (Rate Monotonic)-DU (Decrease Utilization)-NFS
(Next Fit Scheduling) [19] strategy does not suffer from
execution time and period anomalies. Its a scheduling of
periodically arriving tasks on multiprocessor environment.
This strategy partitioning tasks by using a variant of nextfit-bin-packing. If the following assumption does not hold;
(1) assume that a task meet its deadline if it did so when all
tasks are executed at their maximum execution time, or (2)
assume that a task meets its deadline if it did so when all
tasks arrived frequently, then this situation is known as
scheduling anomalies. RM-DU-NFS is the combination of
DU and NFS which is based on high system utilization
bound.
Optimal Preemptive Scheduling [20] is used to
schedule n jobs on, m parallel uniform machine. By
assigning shortest remaining processing time jobs to the
fastest available machine, discounted flow time is
minimized by serving jobs preemption in increasing order
of their remaining processing time. Shortest Remaining
Processing Time on Fastest Machine (SRPT-FM) rule is
also optimal for the discounted flow time criteria. Flow
time is minimized by scheduling jobs according to the
shortest remaining processing time on the fastest machine.
Minimization of the make span is achieved by scheduling
the (LRPT-FM rule) Longest Remaining Processing Time
on Fastest Machine [21].
PTS (Preemption Threshold Scheduling) [23] disable
the preemptions up to a specific priority level, called
preemption threshold. Regular priority assign to the
arriving tasks, and the tasks may preempt only if the
priority of the arriving task is higher than the threshold of
running task.
In FPP (Fixed Preemption Point) scheduling [24]
strategy, a task is assign in a non-preemptive mode, and a
preemption can take place at a specific point of the code
which is known as preemption point. In this way, a task is
divided into subtasks. If the higher priority task is arrive
then preemption is postpone until the next preemption
point. So the tasks cooperate to each other by the
preemption point.

V.

NON-PREEMPTIVE PARTITIONING
STRATEGIES

LSA (Loose Synchronization Algorithms) use nonpreemptive

deterministic
scheduler
to
maintain
multithreaded replica consistency. Its control concurrency
by the leader (mutex) thread, which enforced the leader
dictated order on the execution of their thread (followers)
[13]. OFT (Optimal Finish Time) non preemptive strategy
is known as NP-complete with many uniform processors

[14].LPT (Largest Processing Time First) scheduling is

near optimal to non-preemptive OFT.
MSBC (Multiple Strict Bound Constraints) scheduling
is non-preemptive static strategy for heterogeneous
computing systems. It perform alternative task priority
schedule instead of Heterogeneous Earliest Finish Time
(HEFT) scheme. Its also used to raise the performance of
parallel computing applications on the heterogeneous
platforms due to macro data flow graph implementation
[16].
EDF (Earliest Deadline First) Strategy, schedule
periodic or sporadic tasks on uniprocessor without
preemption and without inserted idle time [25].Periodic
tasks are invoked after a certain time interval while
sporadic tasks are invoked in arbitrary time but within the
limited time constraints. EDF is universal for the set of
periodic or sporadic tasks.
Comparative analysis of various task partitioning and
scheduling strategies is discussed below. This analysis is
highly beneficial to choose the appropriate strategy under
the different requirement and architecture.

VI.

FUTURE WORK

Advancement of the technologies and the architecture is

a key factor for the mapping heuristics of the task
partitioning strategies. A researcher decides his direction of
the research problem according to the existing strategies.
Comparative analyses of the recent scheduling strategies
provide the standard of the existing work in the field of the
parallel computing systems. In the static scheduling the
most challenging direction is to extend DAG scheduling to
the heterogeneous environment. A heterogeneous
computing environment is a collection of heterogeneous
machines consisting of communication protocols, high
speed interconnections, interfaces and efficient parallel
computing tools. One research avenue in the research
direction is that new model and optimized algorithms
should be designed which cover the major fields of the
parallel computing systems in the efficient manners. A
common terminology is used assigning the tasks onto the
different type of the parallel computing architectures.

VII. CONCLUSION
We have studied problems, discussed in the literature
from the last twenty years. The main intension of the paper
is to provide a suitable framework for comparing past work
in the area of distributed parallel computing systems. Ideal
performance of the strategy is depending upon the
requirement and the architecture used for the distributed
parallel processing systems. From the brief discussion of
scheduling strategies its clear that there is no ideal strategy
for all parallel computing system. The fruitful comparative
analysis of strategies provides easiest comparisons of the
existing systems. In this paper, Dynamic, preemptive and
non-preemptive task partitioning and load balancing
strategies are briefly discussed.

373 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 05, May, 2016

Maninderjit Singh Khanna et al

Classification of Task Partitioning and Load Balancing Strategies in Distributed Parallel Computing Systems

TABLE 1
Comparison of Different Task Partitioning Strategies under the various architectures

S.N.

Name of Strategy

Type
of
Strategy
Static

Architecture

Remark

Ref

PY (Papadimitriou and Yannakakis )

Scheduling

Homogeneous

DAG based scheduling.

Not Optimal.

34

LCTD (Linear Clustering with Task

Duplication)

Static

Homogeneous

DAG based scheduling.

Not Optimal

35

EZ (Edge Zeroing)

Static

Homogeneous

DAG based scheduling

33

MCP (Modified Critical Path)

Static

Homogeneous

DAG based scheduling

Optimal Scheduling

31

ETF (Earliest Time First)

Static

Homogeneous

DAG based scheduling.

Optimal Scheduling.

32

Position Scan Task Scheduling (PSTS)

Dynamic

Heterogeneous

It can be used in centralized and

decentralized manner.

14

Evolutionary Task Scheduling

Static and
Dynamic

Heterogeneous

Previous state information decide the

scheduling environment

26

DPS(Dynamic Priority Scheduling)

Dynamic

Heterogeneous

Difference of top level and bottom level

of node decide the priority of the tasks, Its
DAG based scheduling.

27

DLS (Dynamic Level Scheduling)

Dynamic

Heterogeneous

DAG based scheduling. Priority assign to

the tasks on the basis of GDL
(Generalized Dynamic Level) in the ready
list at every scheduling step.

29

10

LMT(Levelized Min Time ) Scheduling

Dynamic

Heterogeneous

DAG based scheduling. During first phase

level sorting is used and in second phase
greedy heuristic is applied for assigning
priority.

30

11

FPS(Fast Preemptive Scheduling)

Preemptive

Scheduling cost is very low

17

12

PTS(Preemptive Task Scheduling)

Preemptive

Heterogeneous
and
Homogeneous
Homogeneous

Cannot be implemented in Heterogeneous

18

13

PDS(Preemptive
Scheduling )

Deterministic

Preemptive

Heterogeneous

Multiple threads, and there is no inter

replica communication

11

14

Optimal Preemptive Scheduling subject

to release date
RM-DU-NFS Scheduling

Preemptive

Homogeneous

Many step process

12

Preemptive

Heterogeneous

Free from scheduling anomalies

19

16

Optimal Preemptive Scheduling with

discount flow time objective

Preemptive

Homogeneous

SRPT-FM rule is also optimal for

discounted flow time criteria

20

17
18
19

SRPT-FM Scheduling
LRPT-FM Scheduling
PTS(Preemptive Threshold Scheduling)

Preemptive
Preemptive
Preemptive

FPP(Fixed
Scheduling

Preemptive

Minimize flow time

Minimize make span
Its used to reduce unnecessary
preemption
Non preemptive task make preemptive
with the help of preemption point in code
of the job

21
21
23

20

Homogeneous
Homogeneous
Real
Time
System
Real
Time
System

21

LSA(Loose Synchronization Algorithm)

Scheduling

Nonpreemptive

Heterogeneous

Its multithreaded replica strategy in which

leader (thread) control to (follower) thread
by using mutex (locking/unlocking).

13

15

Preemption

Point

374 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 05, May, 2016

Maninderjit Singh Khanna et al

Classification of Task Partitioning and Load Balancing Strategies in Distributed Parallel Computing Systems

22

OFT(Optimal Finish Time ) strategy

23

LPT(Largest Processing Time First)

strategy
MSBT(Multiple
Strict
Bound
Constraints) strategy

24

25

EDF(Earliest
Scheduling

Deadline

First)

Nonpreemptive
Nonpreemptive
Non
preemptive,
static
Nonpreemptive

REFERENCES
[1]

[2]

[3]

[4]

[5]

[6]
[7]

[8]
[9]
[10]

[11]

[12]

[13]
[14]

[15]

[16]

[17]

Gary. E. Christensen, MIMD vs. SIMD parallel processing: A case

study in 3D medical image registration, Parallel Computing 24
(9/10) (1998), pp. 13691383.
Feitelson D.G., A Survey of Scheduling in Multiprogrammed
Parallel Systems, Research Report RC 19790 (87657), IBM T.J.
Watson Research Center, August 1997.
Jia-X. Z.; Wei-M. Z.; A DAG-based partitioning-reconfiguring
scheduling algorithm in network of workstations, High
Performance Computing in the Asia-Pacific Region, 2000.
Proceedings. The Fourth International Conference/Exhibition on,
vol.1. 2000, pp.323-324
Andrews J. B. and Polychronopoulos C. D. An analytical approach
to performance/cost modeling of Parallel computers. Journal of
Parallel and Distributed Computing, 12(4):Aug. 1991, pp.343356.
Menasce, D. A., Saha, D., Porto, S. C. D. S.,Almeida, V. A. F., and
Tripathhi, S. K. Static and dynamic processor scheduling
disciplines in heterogeneous parallel architectures. J. Parallel
Distrib. Comput. 28, 1 (July 1995), pp .3-6.
Gajski,D., and Peir, J., Essential Issue in Multiprocessor, IEEE
Computer Vol 18, No.6 ,1985,PP. 1-5.
Menasce, D. A., Porto, S. C.,and Tripathi, S. K.. Static heuristic
processor assignment in heterogeneous message passing
architectures. Int. J. High Speed Computing. 1994, PP.114135.
Shmoys D. B., Wein J., and Williamson D.P., Scheduling parallel
machines on-line. SIAM Journal on Computing, December 1995.
Nelson R. and Towsley D., Comparison of threshold scheduling
policies for multiple server systems, IBM, Research. Report, 1985.
Feitelson D.G. and Rudolph L. Parallel job scheduling: Issues and
approaches. In IPPS95 Workshop: Job Scheduling Strategies for
Parallel Processing SpringerVerlag, Lecture Notes in Computer
Science LNCS 949, 1995, pp.1-3.
Kwok Y. and Ahmad I., Static Scheduling Algorithms for
Allocating Directed Task Graphs to Multiprocessors, ACM
Computing Surveys 31(4) (December 1999) 406471
Yu-Kwong K. and Ishfaq A., Efficient Scheduling of Arbitrary
Task Graphs to Multiprocessors Using a Parallel Genetic
Algorithm,Journal of Parallel and Distributed Computing, 1997,
pp.1-3.
Grewe D.and M. F. OBoyle. A static task partitioning approach for
heterogeneous systems using OpenCL. In CC11, Mar.-Apr. 2011.
Savvas K and Tahar Kechadi, M. Dynamic Task Scheduling in
Computing Cluster Environments, Proceedings of the
ISPDC/Heterogeneous Parallel Computing, IEEE conference, 2004,
pp.121154.
S. Ali, H.J. Siegel, M. Maheswaran, D. Hensgen and S. Ali. Task
Execution Time Modeling for Heterogeneous Computing System.
Proceedings of Heterogeneous Computing Workshop, 2000, pp.
184-199.
Chen H. On the Design of Task Scheduling in the Heterogeneous
Computing Environments. IEEE Pacific Rim Conference on
Communications, Computers and Signal Processing, 2005.
Ahmed M., S.M.H. Chowdhury, M. Hasan, Fast preemptive task
scheduling algorithm for homogeneous and heterogeneous
distributed memory systems, in: Ninth ACIS International
Conference on Software Engineering, Artificial Intelligence,
Networking, and Parallel/Distributed Computing, 2008, pp. 721
726.

Homogeneous

NP-Complete

14

Homogeneous

Its near optimal to non-preemptive OFT

15

Heterogeneous

Its use macro

implementation

graph

16

Real
Time
System

Universal for the set of periodic or

sporadic tasks

25

data

flow

[18] Radulescu A. and A. J. C. van Gemund. On the complexity of list

scheduling algorithms for distributed memory systems. ACM Intl
Conf. on Supercomputing, Rhodes Greece, 1999.
[19] Andersson B., Jonsson J., Preemptive multiprocessor scheduling
anomalies, Proceedings of IPDPS 2002, pages: 12-19, 2002
[20] Pandelis] D. G., Optimal Preemptive Scheduling on Uniform
Machines with Discounted Flow Time Objectives, European
Journal of Operational Research, Vol. 177, No. 1, 2007, pp. 630637.
[21] Gonzalez T., Optimal Mean Finish Time Preemptive Schedules,
Technical Report 220, Computer Science Department,
Pennsylvania State University 1977.
[22] Renan A. S., Romulo S. de O., "A Heterogeneous Preemptive and
Non-preemptive Scheduling Approach for Real-Rime Systems on
Multiprocessors,", 2012 Second Brazilian Conference on Critical
Embedded Systems, 2012, pp.70-75.
[23] Desrochers M., Lenstra J.K., and Savelsbergh M.W.P., A
Classification Scheme for Vehicle Routing and Scheduling
Problems, European Journal of Operational Research, 1990, pp.
320331.
[24] Burns A. Preemptive Priority Based Scheduling: An Appropriate
Engineering Approach. Technical Report YCS 214, University of
York, 1993, pp.12-18.
[25] Jeffay, K.; Stanat, D.F.; Martel, C.U.; On Non-Preemptive
Scheduling of Period and Sporadic Tasks, Real-Time Systems
Symposium, Proceedings. Twelfth, vol., no., 1991, pp.129-139.
[26] Zamfirache, F.;,Zaharie, D., Craciun, C. ,Evolutionary Task
Scheduling in Static and Dynamic Environments Computational
Cybernetics and Technical Informatics , International Joint
Conference on , vol., no., 2010,pp.619-625.
[27] Ahmad, I.; Dhodhi, M.K.; Ul-Mustafa, R.; DPS: Dynamic Priority
Scheduling Heuristic for Heterogeneous Computing Systems,
Computers and Digital Techniques, IEE Proceedings - , vol.145,
no.6, Nov 1998, pp.411-418.
[28] Norman M. G. and Thanisch P. Models of Machines and
Computation for Mapping in Multicomputers. ACM Comput.
Surv. 1993. Pp.263302.
[29] Sih, G.C.; Lee, E.A., A compile-time scheduling heuristic for
interconnection-constrained
heterogeneous
processor
architectures, Parallel and Distributed Systems, IEEE Transactions
on, vol.4, no.2, Feb 1993, pp.175-187.
[30] Iverson M., F. Ozgumer, and Follen G..Parallelizing Existing
Applications in a Distributed Heterogeneous Environment In
Proceedings of the 4th Heterogeneous Computing Workshop
(HCW95), 1995, pp. 9199.
[31] Wu, M.Y. and Gajski, D. D., Hypertool: A Programming Aid for
Message-Passing Systems. IEEE Transaction Parallel Distributed.
Systems. 1990 pp. 331344.
[32] Hwang, J.-J., Chow, Y.-C., Anger, F. D., and Lee, C.Y.
Scheduling Precedence Graphs in Systems with Inter-processor
Communication Times. SIAM J. Computer, 1989, pp. 245256.
[33] Sarkar, V. Partitioning and Scheduling Parallel Programs for
Multiprocessors. MIT Press, Cambridge, MA. 1989.
[34] Papadimitriou, C. H. and Yannakakis, M., Towards an
Architecture-Independent Analysis of Parallel Algorithms. SIAM
J. Computer, 1990, pp. 324327.
[35] Chen, H., Shirazi, B., and Marquis, J. Performance Evaluation of a
Novel Scheduling Method: Linear Clustering with Task
Duplication. In Proceedings of the 2nd International Conference
on Parallel and Distributed Systems, 1993, pp. 271275.

375 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 05, May, 2016