Parallel Processing

Prof. Mitul K. Patel

Assistant Professor, Computer Engineering Department

Shree Swami Atmanand Saraswati Institute of Technology, Surat

January 2013

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

1 / 135

Outline
1
2

Introduction to subject
Parallel Programming Platforms
Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

2 / 135

Introduction to subject

Introduction to subject

Subject Name: Parallel Processing

Subject Code: 180702

Evaluation Scheme:

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

3 / 135

Introduction to subject

Reference Books
Text Books:
1

Introduction to Parallel Computing, Ananth Grama, Anshul Gupta,

George Karypis, Vipin Kumar, By Pearson Publication

Introduction to Parallel Processing, M. SasiKumar, Dinesh Shikhare,

P.Raviprakash By PHI Publication

Reference Books:
1

Introduction To Parallel Programming - By Steven Brawer

Introduction To Parallel Processing By M.Sasikumar, Dinesh Shikhare

And P. Ravi Prakash

Parallel Computers Architecture And Programming By V. Rajaraman

And C. Siva Ram Murthy

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

4 / 135

Parallel Programming Platforms

Topic Overview

Implicit Parallelism: Trends in Microprocessor Architectures

Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

5 / 135

Parallel Programming Platforms

Scope of Parallelism

Conventional architectures coarsely comprise of a processor, memory

system, and the datapath.
Each of these components present significant performance bottlenecks.
Parallelism addresses each of these components in significant ways.
Different applications utilize different aspects of parallelism -e.g., data
itensive applications utilize high aggregate throughput, server applications utilize high aggregate network bandwidth, and scientific applications typically utilize high processing and memory system performance.
It is important to understand each of these performance bottlenecks.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

6 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

7 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Implicit Parallelism: Trends in Microprocessor

Architectures

Microprocessor clock speeds have posted impressive gains over the past
two decades (two to three orders of magnitude).
Higher levels of device integration have made available a large number
of transistors.
The question of how best to utilize these resources is an important one.
Current processors use these resources in multiple functional units and
execute multiple instructions in the same cycle.
The precise manner in which these instructions are selected and executed provides impressive diversity in architectures.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

8 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Pipelining and Superscalar Execution

Pipelining overlaps various stages of instruction execution to achieve

performance.
At a high level of abstraction, an instruction can be executed while the
next one is being decoded and the next one is being fetched.
This is akin to an assembly line for manufacture of cars.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

9 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Pipelining and Superscalar Execution

Pipelining, however, has several limitations.

The speed of a pipeline is eventually limited by the slowest stage.
For this reason, conventional processors rely on very deep pipelines (20
stage pipelines in state-of-the-art Pentium processors).
However, in typical program traces, every 5-6th instruction is a conditional jump! This requires very accurate branch prediction.
The penalty of a misprediction grows with the depth of the pipeline,
since a larger number of instructions will have to be flushed.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

10 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Pipelining and Superscalar Execution

One simple way of alleviating these bottlenecks is to use multiple

pipelines.
The question then becomes one of selecting these instructions.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

11 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Superscalar Execution: An Example

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

12 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Superscalar Execution: An Example

In the above example, there is some wastage of resources due to data

dependencies.
The example also illustrates that different instruction mixes with identical semantics can take significantly different execution time.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

13 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Superscalar Execution
Scheduling of instructions is determined by a number of factors:
True Data Dependency: The result of one operation is an input to the
next.
Resource Dependency: Two operations require the same resource.
Branch Dependency: Scheduling instructions across conditional branch
statements cannot be done deterministically a-priori.
The scheduler, a piece of hardware looks at a large number of instructions in an instruction queue and selects appropriate number of
instructions to execute concurrently based on these factors.
The complexity of this hardware is an important constraint on superscalar processors.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

14 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Superscalar Execution: Issue Mechanisms

In the simpler model, instructions can be issued only in the order in

which they are encountered. That is, if the second instruction cannot
be issued because it has a data dependency with the first, only one
instruction is issued in the cycle. This is called in-order issue.
In a more aggressive model, instructions can be issued out of order.
In this case, if the second instruction has data dependencies with the
first, but the third instruction does not, the first and third instructions
can be co-scheduled. This is also called dynamic issue.
Performance of in-order issue is generally limited.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

15 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Superscalar Execution: Efficiency Considerations

Not all functional units can be kept busy at all times.

If during a cycle, no functional units are utilized, this is referred to as
vertical waste.
If during a cycle, only some of the functional units are utilized, this is
referred to as horizontal waste.
Due to limited parallism in typical instruction traces, dependencies, or
the inability of the scheduler to extract parallelism, the performance of
superscalar processors is eventually limited.
Conventional microprocessors typically support four-way superscalar execution.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

16 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Very Long Instruction Word (VLIW) Processors

The hardware cost and complexity of the superscalar scheduler is a

major consideration in processor design.
To address this issues, VLIW processors rely on compile time analysis to identify and bundle together instructions that can be executed
concurrently.
These instructions are packed and dispatched together, and thus the
name very long instruction word.
This concept was used with some commercial success in the Multiflow
Trace machine (circa 1984).
Variants of this concept are employed in the Intel IA64 processors.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

17 / 135

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures

Very Long Instruction Word (VLIW) Processors:

Considerations
Issue hardware is simpler.
Compiler has a bigger context from which to select co-scheduled instructions.
Compilers, however, do not have runtime information such as cache
misses. Scheduling is, therefore, inherently conservative.
Branch and memory prediction is more difficult.
VLIW performance is highly dependent on the compiler. A number of
techniques such as loop unrolling, speculative execution, branch prediction are critical.
Typical VLIW processors are limited to 4-way to 8-way parallelism.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

18 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

19 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Limitations of Memory System Performance

Memory system, and not processor speed, is often the bottleneck for
many applications.
Memory system performance is largely captured by two parameters,
latency and bandwidth.
Latency is the time from the issue of a memory request to the time the
data is available at the processor.
Bandwidth is the rate at which data can be pumped to the processor
by the memory system.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

20 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Memory System Performance: Bandwidth and Latency

It is very important to understand the difference between latency and

bandwidth.
Consider the example of a fire-hose. If the water comes out of the hose
two seconds after the hydrant is turned on, the latency of the system
is two seconds.
Once the water starts flowing, if the hydrant delivers water at the rate
of 5 gallons/second, the bandwidth of the system is 5 gallons/second.
If you want immediate response from the hydrant, it is important to
reduce latency.
If you want to fight big fires, you want high bandwidth.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

21 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Memory Latency: An Example

Consider a processor operating at 1 GHz (1 ns clock) connected to a DRAM

with a latency of 100 ns (no caches). Assume that the processor has two
multiply-add units and is capable of executing four instructions in each cycle
of 1 ns. The following observations follow:
The peak processor rating is 4 GFLOPS.
Since the memory latency is equal to 100 cycles and block size is one
word, every time a memory request is made, the processor must wait
100 cycles before it can process the data.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

22 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Memory Latency: An Example

On the above architecture, consider the problem of computing a dot-product

of two vectors.
A dot-product computation performs one multiply-add on a single pair
of vector elements, i.e., each floating point operation requires one data
fetch.
It follows that the peak speed of this computation is limited to one
floating point operation every 100 ns, or a speed of 10 MFLOPS, a
very small fraction of the peak processor rating!

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

23 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Improving Effective Memory Latency Using Caches

Caches are small and fast memory elements between the processor and
DRAM.
This memory acts as a low-latency high-bandwidth storage.
If a piece of data is repeatedly used, the effective latency of this memory
system can be reduced by the cache.
The fraction of data references satisfied by the cache is called the cache
hit ratio of the computation on the system.
Cache hit ratio achieved by a code on a memory system often determines its performance.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

24 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Caches: Example

Consider the architecture from the previous example. In this case, we introduce a cache of size 32 KB with a latency of 1 ns or one cycle. We use this
setup to multiply two matrices A and B of dimensions 32 32. We have
carefully chosen these numbers so that the cache is large enough to store
matrices A and B, as well as the result matrix C .

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

25 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Caches: Example (continued)

The following observations can be made about the problem:

Fetching the two matrices into the cache corresponds to fetching 2K
words, which takes approximately 200 s.
Multiplying two n n matrices takes 2n3 operations. For our problem,
this corresponds to 64K operations, which can be performed in 16K
cycles (or 16 s) at four instructions per cycle.
The total time for the computation is therefore approximately the sum
of time for load/store operations and the time for the computation
itself, i.e., 200 + 16 s.

This corresponds to a peak computation rate of 64K/216 or 303 MFLOPS.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

26 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Caches

Repeated references to the same data item correspond to temporal

locality.
In our example, we had O(n2 ) data accesses and O(n3 ) computation.
This asymptotic difference makes the above example particularly desirable for caches.
Data reuse is critical for cache performance.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

27 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth

Memory bandwidth is determined by the bandwidth of the memory bus

as well as the memory units.
Memory bandwidth can be improved by increasing the size of memory
blocks.
The underlying system takes l time units (where l is the latency of the
system) to deliver b units of data (where b is the block size).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

28 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth: Example

Consider the same setup as before, except in this case, the block size is 4
words instead of 1 word. We repeat the dot-product computation in this
scenario:
Assuming that the vectors are laid out linearly in memory, eight FLOPs
(four multiply-adds) can be performed in 200 cycles.
This is because a single memory access fetches four consecutive words
in the vector.
Therefore, two accesses can fetch four elements of each of the vectors. This corresponds to a FLOP every 25 ns, for a peak speed of 40
MFLOPS.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

29 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth

It is important to note that increasing block size does not change latency of the system.
Physically, the scenario illustrated here can be viewed as a wide data
bus (4 words or 128 bits) connected to multiple memory banks.
In practice, such wide buses are expensive to construct.
In a more practical system, consecutive words are sent on the memory
bus on subsequent bus cycles after the first word is retrieved.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

30 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth

The above examples clearly illustrate how increased bandwidth results

in higher peak computation rates.
The data layouts were assumed to be such that consecutive data words
in memory were used by successive instructions (spatial locality of reference).
If we take a data-layout centric view, computations must be reordered
to enhance spatial locality of reference.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

31 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth: Example

Consider the following code fragment:

for (i = 0; i < 1000; i++)
column sum[i] = 0.0;
for (j = 0; j < 1000; j++)
column sum[i] += b[j][i];
The code fragment sums columns of the matrix b into a vector column sum.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

32 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth: Example

The vector column sum is small and easily fits into the cache
The matrix b is accessed in a column order.
The strided access results in very poor performance.

Figure : Multiplying a matrix with a vector: (a) multiplying column-by-column,

keeping a running sum; (b) computing each element of the result as a dot
product of a row of the matrix with the vector.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

33 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Impact of Memory Bandwidth: Example

We can fix the above code as follows:

for (i = 0; i < 1000; i++)
column sum[i] = 0.0;
for (j = 0; j < 1000; j++)
for (i = 0; i < 1000; i++)
column sum[i] += b[j][i];
In this case, the matrix is traversed in a row-order and performance can be
expected to be significantly better.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

34 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Memory System Performance: Summary

The series of examples presented in this section illustrate the following concepts:
Exploiting spatial and temporal locality in applications is critical for
amortizing memory latency and increasing effective memory bandwidth.
The ratio of the number of operations to number of memory accesses
is a good indicator of anticipated tolerance to memory bandwidth.
Memory layouts and organizing computation appropriately can make a
significant impact on the spatial and temporal locality.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

35 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Alternate Approaches for Hiding Memory Latency

Consider the problem of browsing the web on a very slow network connection. We deal with the problem in one of three possible ways:
we anticipate which pages we are going to browse ahead of time and
issue requests for them in advance;
we open multiple browsers and access different pages in each browser,
thus while we are waiting for one page to load, we could be reading
others; or
we access a whole bunch of pages in one go amortizing the latency
across various accesses.
The first approach is called prefetching, the second multithreading, and the
third one corresponds to spatial locality in accessing memory words.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

36 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Multithreading for Latency Hiding

A thread is a single stream of control in the flow of a program. We illustrate

threads with a simple example:
for (i = 0; i < n; i++)
c[i] = dot product(get row(a, i), b);
Each dot-product is independent of the other, and therefore represents a
concurrent unit of execution. We can safely rewrite the above code segment
as:
for (i = 0; i < n; i++)
c[i] = create thread(dot product, get row(a, i), b);

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

37 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Multithreading for Latency Hiding: Example

In the code, the first instance of this function accesses a pair of vector
elements and waits for them.
In the meantime, the second instance of this function can access two
other vector elements in the next cycle, and so on.
After l units of time, where l is the latency of the memory system, the
first function instance gets the requested data from memory and can
perform the required computation.
In the next cycle, the data items for the next function instance arrive, and so on. In this way, in every clock cycle, we can perform a
computation.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

38 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Multithreading for Latency Hiding

The execution schedule in the previous example is predicated upon

two assumptions: the memory system is capable of servicing multiple
outstanding requests, and the processor is capable of switching threads
at every cycle.
It also requires the program to have an explicit specification of concurrency in the form of threads.
Machines such as the HEP and Tera rely on multithreaded processors
that can switch the context of execution in every cycle. Consequently,
they are able to hide latency effectively.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

39 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Prefetching for Latency Hiding

Misses on loads cause programs to stall.

Why not advance the loads so that by the time the data is actually
needed, it is already there!
The only drawback is that you might need more space to store advanced
loads.
However, if the advanced loads are overwritten, we are no worse than
before!

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

40 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Tradeoffs of Multithreading and Prefetching

Multithreading and prefetching are critically impacted by the memory bandwidth. Consider the following example:
Consider a computation running on a machine with a 1 GHz clock, 4-word
cache line, single cycle access to the cache, and 100 ns latency to DRAM.
The computation has a cache hit ratio at 1 KB of 25% and at 32 KB of
90%.
Consider two cases: first, a single threaded execution in which the entire
cache is available to the serial context, and second, a multithreaded execution with 32 threads where each thread has a cache residency of 1 KB.
If the computation makes one data request in every cycle of 1 ns, you may
notice that the first scenario requires 400MB/s of memory bandwidth and
the second, 0.75 words/ns or 3GB/s.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

41 / 135

Parallel Programming Platforms

Limitations of Memory System Performance

Tradeoffs of Multithreading and Prefetching

Bandwidth requirements of a multithreaded system may increase very

significantly because of the smaller cache residency of each thread.
Multithreaded systems become bandwidth bound instead of latency
bound.
Multithreading and prefetching only address the latency problem and
may often exacerbate the bandwidth problem.
Multithreading and prefetching also require significantly more hardware
resources in the form of storage.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

42 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

43 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

Dichotomy of Parallel Computing Platforms

An explicitly parallel program must specify concurrency and interaction

between concurrent subtasks.
The former is sometimes also referred to as the control structure and
the latter as the communication model.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

44 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

Control Structure of Parallel Programs

Parallelism can be expressed at various levels of granularity from

instruction level to processes.
Between these extremes exist a range of models, along with corresponding architectural support.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

45 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

Control Structure of Parallel Programs

Processing units in parallel computers either operate under the centralized control of a single control unit or work independently.
If there is a single control unit that dispatches the same instruction to
various processors (that work on different data), the model is referred
to as single instruction stream, multiple data stream (SIMD).
If each processor has its own control control unit, each processor can
execute different instructions on different data items. This model is
called multiple instruction stream, multiple data stream (MIMD).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

46 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

SIMD and MIMD Processors

Figure : SIMD architecture (a) and a typical MIMD architecture (b).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

47 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

SIMD Processors

Some of the earliest parallel computers such as the Illiac IV, MPP, DAP,
CM-2, and MasPar MP-1 belonged to this class of machines.
Variants of this concept have found use in co-processing units such as
the MMX units in Intel processors and DSP chips such as the Sharc.
SIMD relies on the regular structure of computations (such as those in
image processing).
It is often necessary to selectively turn off operations on certain data
items. For this reason, most SIMD programming paradigms allow for
an activity mask, which determines if a processor should participate
in a computation or not.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

48 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

Conditional Execution in SIMD Processors

Figure : Executing a conditional statement on an SIMD computer with four

processors: (a) the conditional statement; (b) the execution of the statement in
two steps.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

49 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

MIMD Processors

In contrast to SIMD processors, MIMD processors can execute different

programs on different processors.
A variant of this, called single program multiple data streams (SPMD)
executes the same program on different processors.
It is easy to see that SPMD and MIMD are closely related in terms of
programming flexibility and underlying architectural support.
Examples of such platforms include current generation Sun Ultra Servers,
SGI Origin Servers, multiprocessor PCs, workstation clusters, and the
IBM SP.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

50 / 135

Parallel Programming Platforms

Dichotomy of Parallel Computing Platforms

SIMD-MIMD Comparison

SIMD computers require less hardware than MIMD computers (single

control unit).
However, since SIMD processors are specially designed, they tend to
be expensive and have long design cycles.
Not all applications are naturally suited to SIMD processors.
In contrast, platforms supporting the SPMD paradigm can be built
from inexpensive off-the-shelf components with relatively little effort in
a short amount of time.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

51 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

52 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Communication Model of Parallel Platforms

There are two primary forms of data exchange between parallel tasks
accessing a shared data space and exchanging messages.
Platforms that provide a shared data space are called shared-addressspace machines or multiprocessors.
Platforms that support messaging are also called message passing platforms or multicomputers.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

53 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Shared-Address-Space Platforms

Part (or all) of the memory is accessible to all processors.

Processors interact by modifying data objects stored in this sharedaddress-space.
If the time taken by a processor to access any memory word in the
system global or local is identical, the platform is classified as a uniform
memory access (UMA), else, a non-uniform memory access (NUMA)
machine.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

54 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

NUMA and UMA Shared-Address-Space Platforms

Figure : Typical shared-address-space architectures: (a) Uniform-memory-access

shared-address-space computer; (b) Uniform-memory-access shared-address-space
computer with caches and memories; (c) Non-uniform-memory-access
shared-address-space computer with local memory only.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

55 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

NUMA and UMA Shared-Address-Space Platforms

The distinction between NUMA and UMA platforms is important from
the point of view of algorithm design. NUMA machines require locality
from underlying algorithms for performance.
Programming these platforms is easier since reads and writes are implicitly visible to other processors.
However, read-write data to shared data must be coordinated (this will
be discussed in greater detail when we talk about threads programming).
Caches in such machines require coordinated access to multiple copies.
This leads to the cache coherence problem.
A weaker model of these machines provides an address map, but not
coordinated access. These models are called non cache coherent shared
address space machines.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

56 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Shared-Address-Space vs. Shared Memory Machines

It is important to note the difference between the terms shared address

space and shared memory.
We refer to the former as a programming abstraction and to the latter
as a physical machine attribute.
It is possible to provide a shared address space using a physically distributed memory.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

57 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Message-Passing Platforms

These platforms comprise of a set of processors and their own (exclusive) memory.
Instances of such a view come naturally from clustered workstations
and non-shared-address-space multicomputers.
These platforms are programmed using (variants of) send and receive
primitives.
Libraries such as MPI and PVM provide such primitives.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

58 / 135

Parallel Programming Platforms

Communication Model of Parallel Platforms

Message Passing vs. Shared Address Space Platforms

Message passing requires little hardware support, other than a network.

Shared address space platforms can easily emulate message passing.
The reverse is more difficult to do (in an efficient manner).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

59 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

60 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Physical Organization of Parallel Platforms

We begin this discussion with an ideal parallel machine called Parallel Random Access Machine, or PRAM.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

61 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Architecture of an Ideal Parallel Computer

A natural extension of the Random Access Machine (RAM) serial architecture is the Parallel Random Access Machine, or PRAM.
PRAMs consist of p processors and a global memory of unbounded size
that is uniformly accessible to all processors.
Processors share a common clock but may execute different instructions
in each cycle.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

62 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Architecture of an Ideal Parallel Computer

Depending on how simultaneous memory accesses are handled, PRAMs can

be divided into four subclasses.
Exclusive-read, exclusive-write (EREW) PRAM.
Concurrent-read, exclusive-write (CREW) PRAM.
Exclusive-read, concurrent-write (ERCW) PRAM.
Concurrent-read, concurrent-write (CRCW) PRAM.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

63 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Architecture of an Ideal Parallel Computer

What does concurrent write mean, anyway?

Common: write only if all values are identical.
Arbitrary: write the data from a randomly selected processor.
Priority: follow a predetermined priority order.
Sum: Write the sum of all data items.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

64 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Physical Complexity of an Ideal Parallel Computer

Processors and memories are connected via switches.

Since these switches must operate in O(1) time at the level of words,
for a system of p processors and m words, the switch complexity is
O(mp).
Clearly, for meaningful values of p and m, a true PRAM is not realizable.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

65 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Interconnection Networks for Parallel Computers

Interconnection networks carry data between processors and to memory.

Interconnects are made of switches and links (wires, fiber).
Interconnects are classified as static or dynamic.
Static networks consist of point-to-point communication links among
processing nodes and are also referred to as direct networks.
Dynamic networks are built using switches and communication links.
Dynamic networks are also referred to as indirect networks.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

66 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Static and Dynamic Interconnection Networks

Figure : Classification of interconnection networks: (a) a static network; and (b)

a dynamic network.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

67 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Interconnection Networks

Switches map a fixed number of inputs to outputs.

The total number of ports on a switch is the degree of the switch.
The cost of switch is influenced by the cost of the mapping hardware,
the peripheral hardware and packaging costs.
The cost of mapping hardware grows as the square of the degree of
the switch, the peripheral hardware linearly as the degree, and the
packaging costs linearly as the number of pins.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

68 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Interconnection Networks: Network Interfaces

Processors talk to the network via a network interface.

The network interface may hang off the I/O bus or the memory bus.
In a physical sense, this distinguishes a cluster from a tightly coupled
multicomputer.
The relative speeds of the I/O and memory buses impact the performance of the network.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

69 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies

A variety of network topologies have been proposed and implemented.

These topologies tradeoff performance for cost.
Commercial machines often implement hybrids of multiple topologies
for reasons of packaging, cost, and available components.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

70 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Buses

Some of the simplest and earliest parallel machines used buses.

All processors access a common bus for exchanging data.
The distance between any two nodes is O(1) in a bus. The bus also
provides a convenient broadcast media.
However, the bandwidth of the shared bus is a major bottleneck.
Typical bus based machines are limited to dozens of nodes. Sun Enterprise servers and Intel Pentium based shared-bus multiprocessors are
examples of such architectures.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

71 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Buses

Figure : Bus-based interconnects (a) with no local caches; (b) with local
memory/caches.

Since much of the data accessed by processors is local to the processor, a

local memory can improve the performance of bus-based machines.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

72 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Crossbars

A crossbar network uses an p m grid of switches to connect p inputs to
m outputs in a non-blocking manner.

Figure : A completely non-blocking crossbar network connecting p processors to

crossbar network b memory banks.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

73 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Crossbars

The cost of a crossbar of p processors grows as (p 2 ).

This is generally difficult to scale for large values of p.
Examples of machines that employ crossbars include the Sun Ultra HPC
10000 and the Fujitsu VPP500.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

74 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Networks

Crossbars have excellent performance scalability but poor cost scalability.

Buses have excellent cost scalability, but poor performance scalability.
Multistage interconnects strike a compromise between these extremes.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

75 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Networks

Figure : The schematic of a typical multistage interconnection network.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

76 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network

One of the most commonly used multistage interconnects is the Omega

network.
This network consists of log p stages, where p is the number of inputs/outputs.
At each stage, input i is connected to output j if:

2i,
0 i p/2 1
j=
2i + 1 p, p/2 i p 1

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

77 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network

Each stage of the Omega network implements a perfect shuffle as follows:

Figure : A perfect shuffle interconnection for eight inputs and outputs.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

78 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network

The perfect shuffle patterns are connected using 2 2 switches.
The switches operate in two modes crossover or passthrough.

Figure : Two switching configurations of the 2 2 switch: (a) Pass-through;

(b) Cross-over.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

79 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network

A complete Omega network with the perfect shuffle interconnects and switches
can now be illustrated:

Figure : A complete omega network connecting eight inputs and eight outputs.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

80 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network Routing

Let s be the binary representation of the source and t be that of the

destination processor.
The data traverses the link to the first switching node. If the most
significant bits of s and t are the same, then the data is routed in
pass-through mode by the switch else, it switches to crossover.
This process is repeated for each of the log p switching stages.
Note that this is not a non-blocking switch.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

81 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Multistage Omega Network Routing

Figure : An example of blocking in omega network: one of the messages (010 to

111 or 110 to 100) is blocked at link AB.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

82 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Completely Connected Network

Each processor is connected to every other processor.

The number of links in the network scales as O(p 2 ).
While the performance scales very well, the hardware complexity is not
realizable for large values of p.
In this sense, these networks are static counterparts of crossbars.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

83 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Completely Connected and Star

Connected Networks
Example of an 8-node completely connected network.

Figure : (a) A completely-connected network of eight nodes; (b) a Star connected

network of nine nodes.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

84 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Star Connected Network

Every node is connected only to a common node at the center.

Distance between any pair of nodes is O(1). However, the central node
becomes a bottleneck.
In this sense, star connected networks are static counterparts of buses.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

85 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Linear Arrays, Meshes, and k-d

Meshes

In a linear array, each node has two neighbors, one to its left and one
to its right. If the nodes at either end are connected, we refer to it as
a 1-D torus or a ring.
A generalization to 2 dimensions has nodes with 4 neighbors, to the
north, south, east, and west.
A further generalization to d dimensions has nodes with 2d neighbors.
A special case of a d-dimensional mesh is a hypercube. Here, d = log p,
where p is the total number of nodes.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

86 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Linear Arrays

Figure : Linear arrays: (a) with no wraparound links; (b) with wraparound link.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

87 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Two- and Three Dimensional Meshes

Figure : Two and three dimensional meshes: (a) 2-D mesh with no wraparound;
(b) 2-D mesh with wraparound link (2-D torus); and (c) a 3-D mesh with no
wraparound.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

88 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Hypercubes and their Construction

Figure : Construction of hypercubes from hypercubes of lower dimension.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

89 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Properties of Hypercubes

The distance between any two nodes is at most log p.

Each node has log p neighbors.
The distance between two nodes is given by the number of bit positions
at which the two nodes differ.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

90 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Tree-Based Networks

Figure : Complete binary tree networks: (a) a static tree network; and (b) a
dynamic tree network.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

91 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Tree Properties

The distance between any two nodes is no more than 2 log p.

Links higher up the tree potentially carry more traffic than those at the
lower levels.
For this reason, a variant called a fat-tree, fattens the links as we go
up the tree.
Trees can be laid out in 2D with no wire crossings. This is an attractive
property of trees.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

92 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Network Topologies: Fat Trees

Figure : A fat tree network of 16 processing nodes.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

93 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Evaluating Static Interconnection Networks

Diameter: The distance between the farthest two nodes in the network. The diameter

of a linear array is p 1, that of a mesh is 2( p 1), that of a tree and hypercube

is log p, and that of a completely connected network is O(1).
Bisection Width: The minimum number of wires you must cut to divide the network
into two equal parts. The bisection width of a linear array and tree is 1, that of a

mesh is p, that of a hypercube is p/2 and that of a completely connected network

2
is p /4.
The number of bits that can be communicated simultaneously over a link connecting
two nodes is called the Channel Width.
The peak rate at which a single physical wire can deliver bits is called the Channel
Rate.
The peak rate at which data can be communicated between the ends of communication link is called Channel Bandwidth.
The Bisection bandwidth of a network is defined as the minimum volume of communication allowed between any two halves of the network. It is also sometimes
referred to as Cross-section bandwidth.
Cost: The number of links or switches (whichever is asymptotically higher) is a
meaningful measure of the cost. However, a number of other factors, such as the
ability to lay out the network, the length of wires, etc., also factor in to the cost.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

94 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Evaluating Static Interconnection Networks

Network

Diameter

Bisection
Width

Arc
Connectivity

Cost
(No. of links)

Completely-connected
Star
Complete binary tree
Linear array
2-D mesh, no wraparound
2-D wraparound mesh
Hypercube
Wraparound k-ary d-cube

1
2
2 log((p + 1)/2)
p1

2( p 1)

2b p/2c
log p
dbk/2c

p 2 /4
1
1
1

2 p
p/2
2k d1

p1
1
1
1
2
4
log p
2d

p(p 1)/2
p1
p1
p1

2(p p)
2p
(p log p)/2
dp

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

95 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Evaluating Dynamic Interconnection Networks

Network
Crossbar
Omega Network
Dynamic Tree

Prepared By: Prof. Mitul K. Patel

Diameter
1
log p
2 log p

Bisection
Width
p
p/2
1

Parallel Processing

Arc
Connectivity
1
2
2

Cost
(No. of links)
p2
p/2
p1

January 2013

96 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Cache Coherence in Multiprocessor Systems

Interconnects provide basic mechanisms for data transfer.

In the case of shared address space machines, additional hardware is
required to coordinate access to data that might have multiple copies
in the network.
The underlying technique must provide some guarantees on the semantics.
This guarantee is generally one of serializability, i.e., there exists some
serial order of instruction execution that corresponds to the parallel
schedule.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

97 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Cache Coherence in Multiprocessor Systems

When the value of a variable is changes, all its copies must either be invalidated or updated.

Figure : Cache coherence in multiprocessor systems: (a) Invalidate protocol; (b)

Update protocol for shared variables.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

98 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Cache Coherence: Update and Invalidate Protocols

If a processor just reads a value once and does not need it again, an
update protocol may generate significant overhead.
If two processors make interleaved test and updates to a variable, an
update protocol is better.
Both protocols suffer from false sharing overheads (two words that are
not shared, however, they lie on the same cache line).
False sharing refers to the situation in which different processors update
different parts of the same cache-line. Thus, although the updates are
not performed on shared variables, the system does not detect this.
Most current machines use invalidate protocols.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

99 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Maintaining Coherence Using Invalidate Protocols

Each copy of a data item is associated with a state.

One example of such a set of states is, shared, invalid, or dirty.
In shared state, there are multiple valid copies of the data item (and
therefore, an invalidate would have to be generated on an update).
In dirty state, only one copy exists and therefore, no invalidates need
to be generated.
In invalid state, the data copy is invalid, therefore, a read generates a
data request (and associated state changes).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

100 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Maintaining Coherence Using Invalidate Protocols

Figure : State diagram of a simple three-state coherence protocol.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

101 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Maintaining Coherence Using Invalidate Protocols

Figure : Example of parallel program execution with the simple three-state

coherence protocol
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

102 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Snoopy Cache Systems

How are invalidates sent to the right processors?
In snoopy caches, there is a broadcast media that listens to all invalidates
and read requests and performs appropriate coherence operations locally.

Figure : A simple snoopy bus based cache coherence system.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

103 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Performance of Snoopy Caches

Once copies of data are tagged dirty, all subsequent operations can be
performed locally on the cache without generating external traffic.
If a data item is read by a number of processors, it transitions to the
shared state in the cache and all subsequent read operations become
local.
If processors read and update data at the same time, they generate
coherence requests on the bus which is ultimately bandwidth limited.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

104 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Directory Based Systems

In snoopy caches, each coherence operation is sent to all processors.

This is an inherent limitation.
Why not send coherence requests to only those processors that need
to be notified?
This is done using a directory, which maintains a presence vector for
each data item (cache line) along with its global state.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

105 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Directory Based Systems

Figure : Architecture of typical directory based systems: (a) a centralized

directory; and (b) a distributed directory.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

106 / 135

Parallel Programming Platforms

Physical Organization of Parallel Platforms

Performance of Directory Based Schemes

The need for a broadcast media is replaced by the directory.

The additional bits to store the directory may add significant overhead.
The underlying network must be able to carry all the coherence requests.
The directory is a point of contention, therefore, distributed directory
schemes must be used.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

107 / 135

Parallel Programming Platforms

Communication Costs in Parallel Machines

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

108 / 135

Parallel Programming Platforms

Communication Costs in Parallel Machines

Communication Costs in Parallel Machines

Along with idling and contention, communication is a major overhead

in parallel programs.
The cost of communication is dependent on a variety of features including the programming model semantics, the network topology, data
handling and routing, and associated software protocols.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

109 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

110 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Message Passing Costs in Parallel Computers

The total time to transfer a message over a network comprises of the following:
Startup time (ts ): Time spent at sending and receiving nodes (executing the routing algorithm, programming routers, etc.).
Per-hop time (th ): This time is a function of number of hops and
includes factors such as switch latencies, network delays, etc.
Per-word transfer time (tw ): This time includes all overheads that are
determined by the length of the message. This includes bandwidth of
links, error checking and correction, etc.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

111 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Store-and-Forward Routing

A message traversing multiple hops is completely received at an intermediate hop before being forwarded to the next hop.
The total communication cost for a message of size m words to traverse
l communication links is
tcomm = ts + (mtw + th )l.

(1)

In most platforms, th is small and the above expression can be approximated by

tcomm = ts + mltw .

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

112 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Routing Techniques

Figure : Passing a message from node P0 to P3 (a) through a store-and-forward

communication network; (b) and (c) extending the concept to cut-through
routing. The shaded regions represent the time that the message is in transit.
The startup time associated with this message transfer is assumed to be zero.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

113 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Packet Routing
Store-and-forward makes poor use of communication resources.
Packet routing breaks messages into packets and pipelines them through
the network.
Since packets may take different paths, each packet must carry routing information, error checking, sequencing, and other related header
information.
The total communication time for packet routing is approximated by:
tcomm = ts + th l + tw m
where
tw = tw 1 + tw 2 (1 + s/r )
The factor tw accounts for overheads in packet headers.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

114 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Cut-Through Routing

Takes the concept of packet routing to an extreme by further dividing

messages into basic units called flits.
Since flits are typically small, the header information must be minimized.
This is done by forcing all flits to take the same path, in sequence.
A tracer message first programs all intermediate routers. All flits then
take the same route.
Error checks are performed on the entire message, as opposed to flits.
No sequence numbers are needed.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

115 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Cut-Through Routing

The total communication time for cut-through routing is approximated

by:
tcomm = ts + th l + tw m.
This is identical to packet routing, however, tw is typically much smaller.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

116 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Simplified Cost Model for Communicating Messages

The cost of communicating a message between two nodes l hops away
using cut-through routing is given by
tcomm = ts + lth + tw m.
In this expression, th is typically smaller than ts and tw . For this
reason, the second term in the RHS does not show, particularly, when
m is large.
Furthermore, it is often not possible to control routing and placement
of tasks.
For these reasons, we can approximate the cost of message transfer by
tcomm = ts + tw m.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

117 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Simplified Cost Model for Communicating Messages

It is important to note that the original expression for communication

time is valid for only uncongested networks.
If a link takes multiple messages, the corresponding tw term must be
scaled up by the number of messages.
Different communication patterns congest different networks to varying
extents.
It is important to understand and account for this in the communication
time accordingly.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

118 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Cost Models for Shared Address Space Machines

While the basic messaging cost applies to these machines as well, a

number of other factors make accurate cost modeling more difficult.
Memory layout is typically determined by the system.
Finite cache sizes can result in cache thrashing.
Overheads associated with invalidate and update operations are difficult
to quantify.
Spatial locality is difficult to model.
Prefetching can play a role in reducing the overhead associated with
data access.
False sharing and contention are difficult to model.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

119 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Routing Mechanisms for Interconnection Networks

How does one compute the route that a message takes from source to
destination?
Routing must prevent deadlocks for this reason, we use dimensionordered or e-cube routing. Dimension ordered routing assigns successive channels for traversal by message based on numbering scheme
determined by the dimension of the channel.
Routing must avoid hot-spots for this reason, two-step routing is
often used. In this case, a message from source s to destination d
is first sent to a randomly chosen intermediate processor i and then
forwarded to destination d.
The dimension ordered routing technique for a two dimensional mesh is
called XY-routing and that for a hypercube is called E-cube routing.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

120 / 135

Parallel Programming Platforms

Messaging Cost Models and Routing Mechanisms

Routing Mechanisms for Interconnection Networks

Figure : Routing a message from node Ps (010) to node Pd (111) in a

three-dimensional hypercube using E-cube routing.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

121 / 135

Parallel Programming Platforms

Mapping Techniques

Outline

Introduction to subject

Parallel Programming Platforms

Implicit Parallelism: Trends in Microprocessor Architectures
Limitations of Memory System Performance
Dichotomy of Parallel Computing Platforms
Communication Model of Parallel Platforms
Physical Organization of Parallel Platforms
Communication Costs in Parallel Machines
Messaging Cost Models and Routing Mechanisms
Mapping Techniques

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

122 / 135

Parallel Programming Platforms

Mapping Techniques

Mapping Techniques for Graphs

Often, we need to embed a known communication pattern into a given

interconnection topology.
We may have an algorithm designed for one network, which we are
porting to another topology.
For these reasons, it is useful to understand mapping between graphs.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

123 / 135

Parallel Programming Platforms

Mapping Techniques

Mapping Techniques for Graphs: Metrics

When mapping a graph G (V , E ) into G 0 (V 0 , E 0 ), the following metrics are

important:
The maximum number of edges mapped onto any edge in E 0 is called
the congestion of the mapping.
The maximum number of links in E 0 that any edge in E is mapped
onto is called the dilation of the mapping.
The ratio of the number of nodes in the set V 0 to that in set V is
called the expansion of the mapping.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

124 / 135

Parallel Programming Platforms

Mapping Techniques

Impact of process mapping on performance

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

125 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Linear Array into a Hypercube

A linear array (or a ring) composed of 2d nodes (labeled 0 through 2d 1)
can be embedded into a d-dimensional hypercube by mapping node i of the
linear array onto node G (i, d) of the hypercube. The function G (i, x) is
defined as follows:

G (0, 1) = 0
G (1, 1) = 1

G (i, x),
i < 2x
G (i, x + 1) =
x
x+1
2 + G (2
1 i, x), i 2x
The function G is called the binary reflected Gray code (RGC).
Since adjoining entries (G (i, d) and G (i + 1, d)) differ from each other at
only one bit position, corresponding processors are mapped to neighbors
in a hypercube. Therefore, the congestion, dilation, and expansion of the
mapping are all 1.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

126 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Linear Array into a Hypercube: Example

Figure : (a) A three-bit reflected Gray code ring; and (b) its embedding into a
three-dimensional hypercube.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

127 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Mesh into a Hypercube

A 2r 2s wraparound mesh can be mapped to a 2r +s -node hypercube by

mapping node (i, j) of the mesh onto node G (i, r 1)kG (j, s 1) of the
hypercube (where k denotes concatenation of the two Gray codes).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

128 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Mesh into a Hypercube

Figure : (a) A 4 4 mesh illustrating the mapping of mesh nodes to the nodes in
a four-dimensional hypercube; and (b) a 2 4 mesh embedded into a
three-dimensional hypercube.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

129 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Mesh into a Linear Array

Since a mesh has more edges than a linear array, we will not have an
optimal congestion/dilation mapping.
We first examine the mapping of a linear array into a mesh and then
invert this mapping.
This gives us an optimal mapping (in terms of congestion).

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

130 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Mesh into a Linear Array: Example

Figure : (a) Embedding a 16 node linear array into a 2-D mesh; and (b) the
inverse of the mapping. Solid lines correspond to links in the linear array and
normal lines to links in the mesh.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

131 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Hypercube into a 2-D Mesh

Each p node subcube of the hypercube is mapped to a p node row

of the mesh.
This is done by inverting the linear-array to hypercube mapping.
This can be shown to be an optimal mapping.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

132 / 135

Parallel Programming Platforms

Mapping Techniques

Embedding a Hypercube into a 2-D Mesh: Example

Figure : Embedding a hypercube into a 2-D mesh.

Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

133 / 135

Parallel Programming Platforms

Mapping Techniques

Cost-Performance Tradeoffs
If the cost of the network is proportional to the number of wires, then
a square p-node wraparound mesh with (log p)/4 wires per channel
costs as much as a p-node hypercube with one wire per channel.
The average communication latency for a hypercube is given by
ts + th (log p)/2 + tw m
and that for a wraparound mesh for the same cost is

ts + th p/2 + 4tw m/(log p)

For a fixed number of nodes, as the message size is increased, the
communication term due to tw dominates.
With store-and-forward routing, the mesh is no longer more cost-efficient
than a hypercube.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

134 / 135

Parallel Programming Platforms

Mapping Techniques

Cost-Performance Tradeoffs
As the number of messages increases, there is a contention on the network. Contention affects the mesh network more adversely than the hypercube network. Therefore, if the network is heavily loaded, the hypercube will outperform the mesh.
If the cost of a network is proportional to its bisection width, then a p-node wraparound

mesh with p/4 wires per channel has a cost equal to a p-node hypercube with one
wire per channel.
The communication times for the hypercube and the mesh networks of the same
cost are given by
ts + th (log p)/2 + tw m
and

ts + th p/2 + 4tw m/ p

, respectively.
For a large enough messages, a mesh is always better than a hypercube of the same
cost, provided the network is lightly loaded.
Even when the network is heavily loaded, the performance of a mesh is similar to
that of a hypercube of the same cost.
Prepared By: Prof. Mitul K. Patel

Parallel Processing

January 2013

135 / 135