Abstract— To improve the efficiency of computation and
to reduce its execution time becomes necessary for any large
application which consumes huge processing cycle. For this
reason many concepts and programming models are evolved.
Out of those concepts concurrent execution of small units of
the same application gives better performance. This paper
addresses the comparison of efficiency and performance of
computing in multiprocessor and multicomputer environment.
A set of tasks having various sizes is run in normal
programming environment, multiprocessor environment and
multicomputer environment in order to analyze the
performance. The empirical results illustrate that the
performance is optimal in multiprocessor (Cluster) system.
Keywords - Multiprocessor System, Multicomputer System,
Grid Gain, Multi Threading, Middleware, Cluster
I. INTRODUCTION
The complexity of working scenario is increasing now-
days. To reduce the time duration and improve the
efficiency of various applications many programming
models evolved. Multiprocessor system and multicomputer
system are widely used these days. In multiprocessor
systems the system contains more than one processor which
share a common memory block known as shared memory.
Where in multicomputer environment multiple computer
systems are there each having their own processor and
memory and that memory blocks are known as distributed
memory. All the systems are connected with each other in a
centralized network. The multicomputer system may contain
heterogeneous devices which may also effect the program
execution. Applications can be deployed in any of these
environments using some libraries, APIs and middleware.
For deploying an application in multiprocessor system
multi-threaded programming model has been used. Libraries
like OpenMp, JOMP support this concept. OpenMP is
developed in C language and it is compatible with C and
C++. JOMP [1] is developed in Java language. However we
can use Java’s inbuilt thread class and concurrency package
to develop multiprocessor applications. To develop and run
applications in distributed environment MPI [2], MPJ, Grid
Gain [3, 4], Hadoop etc are used. MPI and MPJ are libraries
which are developed in C and Java respectively. Grid Gain
and Hadoop are middleware which take the task and invoke
in the grid environment.
Parallelism can be achieved in loop level through threads
in which threads divide the loop iterations among them and
executed the iterations concurrently [5, 6]. In our
experiment the task is executed using loops. For this we
have used multithreaded model to share the task among
threads and run them concurrently.
For our experiment we have used Thread Class and
Concurrency package (java.util.concurrency) [7, 8, and 9]
for multiprocessor application and Grid Gain Middleware
for multicomputer application.
In our experiment we have executed the task [10, 11] in
cluster (Serial Execution), in hybrid system and distributed
system grid gain as middleware and in again in cluster
(Parallel Execution using Threads). All the programs run in
java virtual machine
®
(Version 6.5)
In section II we have discussed about the concept of
multiprocessor system and multicomputer system. Section
III describes the multi-threading and threaded programming
model. The Grid Gain (Grid Middleware) and working
mechanism of Grid Gain are described in Section IV.
Section V describes the set up of cluster and grid
environment. It also describes the procedure to use
multithreaded programming model. And finally the
comparison of performance between multiprocessor
(Cluster) and multicomputer (Desktop Grid) systems is
described in section VI.
II. MULTIPROCESSOR SYSTEM AND MULTICOMPUTER
SYSTEM
A Multiprocessor System contains more than one fully
programmable processors [12], each capable of executing its
own programs. Multiprocessors are multiple CPU Systems
which access a shared memory. In this system each
processor can access the main memory along with other
peripherals to process programs simultaneously. The
Comparative Analysis of Workflow and Performance Characteristics in Cluster
and Desktop Grid
K Sudipta Achary
1
, Arkaprava Bhaduri Mandal
2
, Motahar Reza
3
School of Computer Science and Engineering
National Institute of Science and Technology
Palur Hills, Berhampur, 761008
1
sudipta.achary@gmail.com,
2
arkabhaduri@gmail.com,
3
reza@nist.edu
978-1-4673-1344-5/12/$31.00 ©2012 IEEE