A Parallel Computing-a Paradigm to achieve High Performance

International Journal of Innovative Research in Advanced Engineering (IJIRAE)
Volume 1 Issue 2 (April 2014)
_______________________________________________________________________________________________
ISSN: 2278-2311 IJIRAE | http://ijirae.com
© 2014, IJIRAE - All Rights Reserved Page - 62
A Parallel Computing-a Paradigm to achieve High
Performance
Prof.Ashwini M. Bhugul
Asst. prof. MMCOE, Pune
ashu_b53@rediffmail.com
Abstract-Over last few years there has been rapid changes found in computing field.today, we are using the latest
upgrade system which provides the faster output and high performance. User view towards computing is only to
get the correct and fast result. There are many techniques which improves the system performance. Today’s
widely use computing method is parallel computing. Parallel computing, including foundational and theoretical
aspects, systems, languages, architectures, tools, and applications. It will address all classes of parallel-
processing platforms including concurrent, multithreaded, multicore, accelerated, multiprocessor, clusters, and
supercomputers. This paper reviews the overview of parallel processing to show how parallel computing can
improve the system performance.
1. INTRODUCTION
Parallel computer architecture provides a proportional increase in performance with increase in performance with
increasing system resources. System resources are scaled by the number of processors used, memory capacity
enlarged, access latency tolerated, I/O bandwidth required, etc.parallel architecture has a higher potential to deliver
scalable performance Parallelism appears in various forms such as look ahead, pipelining, vectorization,
concurrency, data parallelism, etc.
Parallel computers are those that execute programs in MIMD code. There are two classes, namely-shared-
memory multiprocessors and message passing multicomputers.In parallel computing a large task is divided into
subtask and all the sub tasks will execute in parallel. This will help to reduce the waiting or stalled stages in pipeline
of the computer system. Lower the number of clock cycles required higher performance will be achieve by the
system. The degree of parallelism depends upon the parallel computer architecture. A parallel computer (or multiple
processor system) is a collection of communicating processing elements (processors) that cooperate to solve large
computational problems fast by dividing such problems into parallel tasks.
There are some classification made for parallel computing, one of is Flynn’s classification. Flynn’s classification is
based on multiplicity of instruction streams and data streams in computer system. Its machine organization includes
SISD (single instruction single data), SIMD (single instruction multiple data), MISD (multiple instruction single
data), and MIMD (multiple instruction multiple data).
Some Broad issues involved:
1. The concurrency and communication characteristics of parallel algorithms for a given computational problem.
2. What portions of the computation and data are allocated or mapped to each PE.
3. How the processing elements cooperate and communicate.
3. How data is shared/transmitted between processors.
4. Abstractions and primitives for cooperation/communication and synchronization.
Parallel computing improves the system performance, as the complete task is divided into stages, which will take
less time to execute the stream of data. Both data and instructions are fetched by memory module. Instructions are
decoded by the control unit, which sends to processor units for execution. Each instruction stream is generated by an
independent control unit. Multiple data stream originates from shared memory module. This improves the system
speedup and performance.
1.1General overview of parallel processing system:

_______________________________________________________________________________________________
CU-control unit
PU-processing unit
MEM-memory
Figure 1.1: SIMD architecture showing parallel computing
1.2 Need of Parallel Computing
Whenever there is need to work on large amount of data, system need a fast performance, more computing
cycles/memory needed, there will be overload on machine when it performs a whole task, parallel computing make
this task efficient by dividing a task into stages that will reduce the computing clock cycles and memory too. Some
of the applications of parallel computing are: Scientific/Engineering computing: CFD, Biology, Chemistry, Physics,
General-purpose computing: Video, Graphics, CAD, Databases, Transaction Processing, Gaming, etc.
Number of transistors on chip growing rapidly. Clock rates expected to continue to go up but only slowly.
Actual performance returns diminishing due to deeper pipelines. Increased transistor density allows integrating
multiple processor cores per creating Chip-Multiprocessors (CMPs) even for mainstream computing applications
that is also one of application of parallel processing The increased utilization of commodity of-the-shelf (COTS)
components in high performance parallel computing systems instead of costly custom components used in
traditional Super computers leading to much lower parallel system cost. Today’s microprocessors offer high-
performance and have multiprocessor support with parallel processing.
II. DIFFERENT GRANULARITIES OF PARALLELISM
In parallel computing units of work can be execute concurrently if sufficient recourses exists. There are two
granularities:
2.1 Thread level parallelism
2.2Instruction level parallelism
2.1Thread level parallelism (TLP)
A thread is separate process having its own instructions and data. Thread may be a separate process or single
program executing in parallel. In parallel computing, the thread offers different instructions and data so that the
processor can switch to another thread and can continue the instructions, this will include the thing that, processor
will continuously execute instructions without stalls, this will take less time to execute instructions.TLP is a
software that enables a program often a high end programs such as web application, etc.
There are two approaches to TLP:
1. Fine-grained multithreading
2. Coarse-grained multithreading
Fine grained multithreading switches threads on each instruction, that will execute multiple instructions
simultaneously, CPU must be able to switch threads every clock cycles. The advantage of using this technique is
that it can hide both short and long stalls. Coarse-grained multithreading switches threads only when current thread
is likely to stall for some time.it is more natural for any given thread and consume less time to switch threads. It is
also easier to implement switching process.
CU
PU1
PU2
PUn
MEM1
MEM2
MEMn

_______________________________________________________________________________________________
2.2 Instruction Level Parallelism
Instruction-level Parallelism CILP) is a family of processor and compiler design techniques that speed up execution
by causing individual machine operations to execute in parallel. Pipelining can overlap the execution of instructions
when they are independent on each other, this potential overlap among instructions is called as Instruction level
Parallelism. Processing instructions in parallel requires some tasks:
1. Checking independencies between instructions, if two instructions are independent on each other, they can
be group together to execute in parallel.
2. Assigning instructions to the functional units on hardware.
3. Determining when instructions are initiated. This include the technique called score boarding, it allows
instructions to execute out of order when there is sufficient resource available.
ILP architectures can be classified as follows.
 Sequential architectures: architectures for which the program is not expected to convey any explicit
information regarding parallelism. Superscalar processors are representative of ILP processor
implementations for sequential architectures.
 Dependence architectures: This architecture provides information of those programs in which instructions
which are dependent on each other. Dataflow processors are one of the examples of this architecture.
 Independence architectures: architectures for which the program provides information as to which
operations are independent of one another. Very Long Instruction Word (VLIW) processors are examples
of the class of independence architectures.
III. SOME PARALLEL PROGRAMMING TECHNIQUES
Sometimes in large scale industries, there is a multiprocessor system, which requires number of instructions to be
executed simultaneously, as the system consists of number of processor connected together there may happens some
memory conflicts while performing tasks parallel. Processor needs to communicate with each or want to share some
data with each other; hence there are some parallel programming techniques, three of them are explained below:
3.1. Message passing parallel programming
In massage passing programming programs view their programs as a collection of cooperating processes
with private variables. This style is suitable for message passing parallel computers. This technique needs only two
primitives that is send and receive, that will send or receive instructions from one processor to another. Figure given
below showing Message-Passing example of Intel Paragon.
Figure 31: Sandia’s Intel Paragon XP/S-based Supercomputer
There are most widely used Message Passing Programming tools are:
1. Message Passing Interface (MPI):
It provides a standard for writing concurrent message-passing programs.MPI offers rich set of functions for
doing point to point communicaton.it will divide a task into stages and that separate tasks can communicate with
each other.MPI implementations include parallel libraries used by existing programming languages (C, C++).
2. Parallel Virtual Machine (PVM):

_______________________________________________________________________________________________
PVM is portable message passing programming system, designed to form separate host machines to from
“virtual machine”. Virtual machine is collection of processors that may work simultaneously or at different time. It
enables a collection of heterogeneous computers to use as a coherent and flexible concurrent computational
resource. PVM support software executes on each machine in a user configurable pool, and provides computational
environment of concurrent applications. User programs written for example in C, FORTRAN or Java are provided
access to PVM through the use of calls to PVM library routines.
3.2. Shared Memory Programming
In shared memory programming, programmers view their programs as a collection of cooperating processes that will
access a common pool of shared variables. In such computers, processors may not have private program or data
memory. a common data or program is stored in shared memory that will use by all processors. The main program
creates separate processes and assigns a tasks to every processor with a common shared memory, each processor
computes independently and after all, result of each processor stored in common memory. Two statements are use
for this purpose: 1. Fork 2. Join
Example: Process X; Process Y;
fork y;
join y;
end y;
Process X when it encounters fork y invokes another process y, after invoking y, it continues doing its work. Process
y starts executing concurrently in another processor. When X reaches join y statement has to wait till y finish its
work. In shared memory programming when multiple processors are in use and sharing a common memory so care
has to be taken that no two dependent instructions should execute simultaneously. For shared memory programming
three models of multiprocessors are:
1. The Uniform Memory Access (UMA) Model:
In this model all physical memory is shared by all processors. All processors have equal access to all memory
addresses.
2. Distributed memory or Non-uniform Memory Access (NUMA) Model:
In this model, Shared memory is physically distributed locally among processors. Access latency to remote memory
is higher.
3. The Cache-Only Memory Architecture (COMA) Model:
In this model, A special case of a NUMA machine where all distributed main memory is converted to caches. No
memory hierarchy at each processor.
3.3 Data parallel Programming
In Data Parallel Programming many programs can apply same operation to each element of composite
data.sometimes, looping statements performed repetitive tasks, using data parallelism this repetition carried out
automatically by having multiprocessors working concurrently. Logically, single thread of control, performs
sequential or parallel steps conceptually, a processor is associated with each data element.
Figure 3.3: Data parallel system (All PE’s are synchronized)
Control Processor
PE PE PE
PE PE PE
PE PE PE

_______________________________________________________________________________________________
Example machines, Thinking Machines CM-1, Maspar MP-1 and MP-2, etc.Data parallel programming
makes programs easy to read and write. Same program will be executed over different data sets simultaneously on
multiple processors. a popular data parallel programming language is high performance Fortran (HPF). figure 3.3
given below shows example of data parallel systems. Array of many simple processors each with little memory.
Processors don’t sequence through instructions. Attached to a control processor that issues instructions Specialized
and general communication, global synchronization
IV.CONCLUSION
At the end of this review paper, we may observe some current some current trends. High performance
computers are increasingly in demand, Parallel computing architecture is today’s widley used architecture which
provides speed and high performance. It is used in many scientific, engineering, medical, military and research areas
to achieve high system performance.
V.REFERENCES
1. Rafiqul Zaman Khan, Md Firoj A Comparative Study on Parallel Programming Tools in Parallel Distributed
Computing System: MPI and PVM, Department of Computer Science, Aligarh Muslim University, Aligarh-
202002,(U.P), India.
2 Ramakrishna Rau, Joseph A. Fisher, Instruction-Level Parallel Processing: History, Overview and Perspective,
Computer Systems Laboratory,HPL-92-132,October, 1992.
3. M. D. Smith, M. Horowitz and M. Lam. Efficient superscalar performance through boosting. Proc. Fifth
International Conference on Architectural Support for Programming& Laneua&es and Operating Systems
(Boston, Massachussetts, October 1992), 248-259.
4. R. Chaiken, B. Jenkins, P.-_A. Larson,B. Ramsey, D. Shakib, S. Weaver, and J. Zhou. SCOPE: Easy and
e_cient parallel processing of massive data sets. International Conference of Very Large Data Bases (VLDB),
August 2008.
5. D. DeWitt and J. Gray. Parallel database systems: The future of high performance database processing.
Communications of the ACM, 36(6), 1992.
6. R. Pike, S. Dorward, R. Griesemer, and S. Quinlan. Interpreting the data: Parallel analysis with Sawzall.
Scienti_c Programming,13(4):277{298, 2005.
7. F. Rabhi and S. Gorlatch. Patterns and Skeletons for Parallel and Distributed Computing. Springer, 2003.
8. Y. Yu, M. Isard, D. Fetterly, M. Budiu,U. Erlingsson, P. K. Gunda, and J. Currey.DryadLINQ: A system for
general-purpose distributed data-parallel computing using a high-level language. In Proceedings of the 8th
Symposium on Operating Systems Design and Implementation (OSDI), December 8-10 2008.
9. R. Z. Khan, A. Q. Ansari and Kalim Qureshi―Performance Prediction for Parallel Scientific Applicationǁ,
Malaysian Journal of Computer Science,Vol. 17 No. 1, June 2004, pp 65-73.
10. Kai Hwang, Faye A. Briggs, "Computer Architecture and Parallel Processing" McGraw-Hill international
Edition
11. Kai Hwang, "Advanced Computer Architecture", Tata McGraw-Hill
12. V.Rajaraman, L Sivaram Murthy, "Parallel Computers", PHI.

A Parallel Computing-a Paradigm to achieve High Performance

More Related Content

What's hot

Viewers also liked

Similar to A Parallel Computing-a Paradigm to achieve High Performance

More from AM Publications

Recently uploaded

A Parallel Computing-a Paradigm to achieve High Performance