MetaComputing

1
MetaComputing
CEG 4183

• Benson Chan 1813283
• Michel Chicoine 1744063

This report was prepared for Professor L.
Orozco-Barbosa in partial fulfillment of the
requirements for the course ELG/CEG 4183
2
Introduction
3
Motivation

• Under-utilized powerful workstations around the
  world
• Unused processor cycles and resources
• Supercomputers are expensive
• Faster networks

4
MetaComputing

• Network of computers that are separated
  geographically to perform computations.
  (Metasystem)
• Transparent distribution of the workload
• High-speed networks used to connect
  supercomputers, database servers, devices.

5
MetaComputing Issues

• Transparency
• Use of standard cross platform language for
  heterogeneous systems
• Synchronization and Scheduling
• Temporal management of resources
• Load Balancing
• Spread tasks among parallel resources
• Security and robustness
• Protect unauthorized use of the resources and
  ensure availability of the resources. (Policy)

6
Parallel Architectures

• SIMD
• MIMD Shared Memory
• MIMD Distributed Memory

7
MIMD Shared Memory
8
MIMD Distributed Memory
9
Programming Models

• Message Passing
• PVM
• MPI
• PVMPI
• Client/Server
• Code Shipping
• Proxy Computing
• Intelligent Mobile Agents

10
MPI

• Designed for High Performance Message Passing for
  Clusters and SuperComputers
• Two standards 1.1 and 2.0 can be found at
• http//www-unix.mcs.anl.gov/mpi/

11
Coding Example
include "mpi.h"
main( argc, argv )
int argc
char argv

char message20
int myrank
MPI_Status status
MPI_Init( argc, argv )
MPI_Comm_rank( MPI_COMM_WORLD, myrank )
if (myrank 0) / code for process zero
/

strcpy(message,"Hello, there")
MPI_Send(message, strlen(message),
MPI_CHAR, 1, 99, MPI_COMM_WORLD)

else / code for process one
/

MPI_Recv(message, 20, MPI_CHAR, 0, 99,
MPI_COMM_WORLD, status)
printf("received s\n", message)

MPI_Finalize()
12
Distributed System Support

• CORBA
• DCOM
• RMI/IIOP

13
Network Support
Source Metacomputing The Informal
Supercomputer, Mark Baker and Geoffrey Fox
3/11/2002
14
Resource Management Issues

• Site autonomy
• Resources operated and administered by
  different parties with different policies,
• security mechanisms, etc.
• Heterogeneous substrate
• Derived from site autonomydifferent resources
  use different resource management schemes
• Policy extensibility
• Need to easily support changes/additions of
  new policies for different domains
• Co-allocation
• Need for mechanism to co-ordinate/manage
  computations on different resources
• Online control
• Need to handle negotiation of resource
  requirements and availability

15
Resource Management Approach

• Network Batch Queuing Systems
• Designed for single domains (no site autonomy)
• Resource management policy set by user (no policy
  extensibility)

16
Present Applications

• Areas
• 1-Distributed supercomputing
• 2-On-demand computing
• 3-High-throughput computing
• 4-Collaborative computing
• 5-Data-intensive computing
• Examples Virtual Laboratory GIS

17
Research Area Grid Computing

• Globus Project
• Solutions for high-performance, large-scale
  resource distributed systems.
• Coordinated resource sharing and problem solving

18
Future Possibility Web Services

• Linking of web servers to exchange data and
  combine information to offer value services
  through any Net device
• Different languages, different platform
• XML the lingua franca
• Standards SOAP, UDDI, WSDL
• .Net vs. J2EE

19
Web Service Example
20
Summary

• Metacomputing allows the use of the Internet to
  increase computing power
• Complex architecture issues across platforms and
  languages
• Basis of the next wave of computing technologies
• Web Services is the next big hype allows
  organizations to save system integration costs,
  generate new revenue models

21
References (contd)

• 1
• Czajkowski, Karl. Foster, Ian. Karonis, Nicholas.
  Kesselman, Carl. Martin, Stuart. Smith, Warren.
  Tuecke, Steven. Resource Management Architecture
  for Metacomputing Systems. Information Sciences
  Institute, University of Southern California,
  Marina Del Ray CA Mathematics and Computer
  Science Division, Argonne National Library,
  Argonne IL.
• 2
• Kacsuk, Peter. Vajda, Ferenc. Network based
  Distributed Computing (Metacomputing). ERCIM
  1999. http//www.ercim.org/publication/prosp/NdBC.
  pdf
• 3
• Foster, Ian. Kesselman, Carl. Tuecke, Steven.
  Anatomy of the Grid Enabling Scalable Virtual
  Organizations. ttp//www.globus.org/research/paper
  s/anatomy.pdf

22
References (contd)

• 4
• Wolter, Roger. XML Web Services Basics. Microsoft
  Developer Network, Microsoft Corporation.
  December 2001. http//msdn.microsoft.com/library/d
  efault.asp?url/library/en-us/dnwebsrv/html/webser
  vbasics.asp
• 5
• Wong, Wylie. Kane, Margaret. Ricciuti, Mike. The
  New Buzz Web Services try to rise above din.
  CNet News.com, November 8 2001.
  http//news.com.com/2009-1017-275484.html?legacyc
  net
• 6
• Message Passing Interface Standard. Message
  Passing Interface Forum.
• http//www.mpi-forum.org/docs/mpi-11-html/mpi-repo
  rt.html

23
References (contd)

• 7
• Baker, Mark. Fox, Geoffrey. Metacomputing The
  Informal Supercomputer Northeast Parallel
  Architectures Center, Syracuse University,
  Syracuse NY.
• http//www.npac.syr.edu/users/mab/homepage/C
  ornell/lecture1/Slide-Show/
• 8
• Webb, Daren. Wendelborn, Andrew. Maciunas, Kevin.
  Process Networks as a High Level Notation for
  MetaComputing. Department of Computer Science,
  University of Adelaide. Adelaide, Australia.
  http//ipdps.eece.unm.edu/1999/java/webb.pdf

24
Q A
25
Study Questions

• 1)What is the communication language of
• Web Services? XML
• 2)T/F? MetaComputing environments are within
• the same geographic location F
• 3)What is a cluster? A collection of PCs
• and workstations.
• 4)What is the Microsoft-proprietary distributed
• architecture called? DCOM

26
Study Questions (contd)
5) What is the difference between MIMD with
shared memory and distributed memory?
Shared memory is a scheme which all processors of
the MetaComputing environment share one
source of memory (they are assigned a share
of the central memory) whereas in
Distributed Memory, each processor has its
own memory at its own physical location.