Next: About this document
Parallel And Distributed Computing
- CSCI 6175 Seminar in Computer Science
Xiannong Meng
Department of Computer Science
The University of Texas - Pan American
Edinburg, TX 78539-2999
meng@panam.edu
Fall 1997
Outline of the Presentation
- What is parallel computing? What is distributed computing?
- Parallel computing
- Problems
- Tools and national centers
- Trends
- Internet-based computing
- SuperWeb
- Issues and efforts
- Distributed simulation
- Conservative and optimistic approaches
- Web-based simulation
- Further probes
Parallel And Distributed Computing
- Parallel computing partitions a task into subtasks each of which
has the same the computing.
- Parellel computing often involves data parallel (SIMD);
shared memory address space.
- Example: stacking books into bookshelves in a library.
- Distributed computing partitions a task into subtasks each of
which may have a different computing task, collectively they
accomplish a task.
- Distributed computing often assumes autonomous computing
agents; they often do not share memory address space; communications
are done through message passing.
- Example: a library has to have a circulation desk, a
reference librarian, a stacking staff put the books back to the shelves.
each subtask
- The difference between parallel and distributed computing is
blurring.
Parallel Computing
- Grand Challenge Problems: in 1993, NSF and a group of
scientists proposed the Grand Challenge Problems which are the
problems that require tera-flop computing power. The
problems range from environmental and earth science to computational
physics.
Examples include
- large scale environmental model
- large scale structure and galaxy formation
- computational quantum materials
- data analysis and knowledge discovery in geophysical databases
- NSF has founded a number of meta-computer centers, along with
a number of established national laboratories to attack these Grand
Challenge Problems.
- Non-government organizations such as Parallel Tools
Consortium also play important roles in the evolving of parallel
computing technology.
- Major software tools available: PVM (Parallel Virtual
Machine) and MPI (Message Passing Interface)
- Move towards the utilization of computing power available
over the Internet.
Internet-Based Computing
- Estimated more than 30 million computers will be connected to
the Internet by the end of 1997
- Most of the computers are used at most half of the time
- Each individual computer may not be as fast, but the
combined computing power can be enormous
- Based on these observations, some researchers proposed
``SuperWeb'', a global Web-based parallel computing infrastructure
- Key ideas:
- There are three kinds of computers in SuperWeb: hosts,
brokers, and clients.
- Hosts offer their resources (CPU time, memory,
storage) as a commodity by registering with a broker.
- Brokers coordinate the supply and demand for computing
resources.
- Clients are users or computers that need extra
computational resources.
- Hosts register their excessive resources with brokers by
sending messages to a broker. By registering with a broker, the host
becomes a part of the SuperWeb.
- Brokers:
- they are responsible for
- Registering hosts
- Accepting client requests
- Overseeing client/host interaction
- A broker consists of three modules: the interface, the
scheduler, and the accouting module.
- Interface: hosts and clients register through
this interface, e.g. use a form to fill out host/client information.
- Scheduler: assign incoming tasks to available
resources.
- Accounting: keep track of the charges incurred
by clients, and credites gained by hosts.
- Clients request resource through a broker. When the request
is granted, the client contacts the hosts directly for computing
power. When the computing is completed, the client disconnects itself
from the hosts.
- Economic model: the hosts can be rewareded for providing
their spare resources by allowing free use of other's CPU, storage
space, license etc.
- Difficulties in Web-based computing come from two major
sources
- Technically: security (solution: encrypted computing)
and speed of communicaiton
- Socially: why letting others use my spare computing
resources? (solution: commodity exchange, coupon for other items) How
can I trust the foregn programs running on my computer?
Distributed Simulations
One of the interesting applications of Internet-based computing is
distributed simulation.
- Discrete event driven simulation
- Parallel and distributed simulation
- Local Causality Constraint: Events are processed in a
non-decreasing timestamp order.
- Conservative approach: make sure no causality is violated;
- Optimistic approach: each logical process can
advance at its own pace. When causality error is detected, a rollback is
carried out to recover the errors.
- We concentrate on optimistic distributed simulation of discrete
event systems.
- Logical process: they can be on the same processor or
different processor
- Event list: they can be implemented in any priority list
- Distributed event list
- Time Warp
- GVT
- rollback
- anti-event (message)
Web-Based Simulation
- Java and its applets
- Key ideas in Web-based simulation:
- Interactive
- Animation
- Heterogeneous environment
For Further Information
- A.D. Alexandrov, M. Ibel, K.E. Schauser, and C.J. Scheiman,
``SuperWeb: Towards a Global Web-Based Parallel Computing
Infrastructure'', <http://www.cs.ucsb.edu/~schauser/papers/>,
April 1997
- A.D. Alexandrov, M. Ibel, K.E. Schauser, and C.J. Scheiman,
``SuperWeb: Research Issues in Java-Based Global Computing'',
<http://www.cs.ucsb.edu/~schauser/papers/>,
June 1997
- E.H. Page, R.L. Moose Jr., S.P. Griffin, ``Web-Based
Simulation in SimJava Using Remote Method Invocation'', Proceedings of
1997 Winter Simulation Conference, December 1997, Atlanta, GA
- R. McNab and F.W. Howell, ``Using Java for Discrete Event
Simulation'', Proceedings of Twelfth UK Computer and
Telecommunications Performance Engineering Workshop, 1996, University
of Edinburgh
- R. Fujimoto, ``Parallel And Distributed Simulation'',
<http://www.cs.gatech.edu/Computing/Pads/pads-intro.ps>, October
1997
- E. Page, ``A Survey of Web-Based Simulation''
<http://ms.is.org/websim/survey/survey.html>
- P. Fishwick, ``Paul Fishwick Computer Simulation''
<http://www.cise.ufl.edu/~fishwick/>
- P.S. Pacheco, ``Parallell Programming with MPI'',
Morgan Kaufmann Publisher, 1997
- X. Meng, ``CSCI 6356: Parallel Computing Courseware'',
<http://www.cs.panam.edu/~meng/Course/CS6356/>
Next: About this document
Xiannong Meng
Wed Dec 3 09:09:58 CST 1997