Unigroup's March 2001 Meeting Announcement


Topic:
Linux Clustering: Beowulf Clusters

Speaker:
Jozef Skvarcek, Ph.D.
Unix Administrator, Datek Online

Date:
Thursday, March 15, 2001

Location:
The Chase Manhattan Bank
55 Water Street, South Tower
13th Floor Conference Room C

Time:
6:15 - 6:30 PM Registration
6:30 - 6:40 PM Ask the Wizard - Questions, Answers and Current Events
6:40 - 6:50 PM Unigroup Business
6:50 - 9:30 PM Main Presentation


Introduction:

The March 2001 Unigroup meeting is on Linux Clustering. The main presentation will made by Jozef Skvarcek, Ph.D, who is an end user of Beowulf. He has designed Beowulf clusters and has designed and coded applications to make use of them. Beowulf is a freely available open-source add-on to Linux which provides support for multi-node clustering. With such a cluster, a group of inexpensive machines can appear to act as a single much larger system. Depending on the size of the workload, the design of the application, and the horsepower and number of cluster members, mainframe computing power can be achieved.

We also have another guest speaker, Andrew Flesch from TurboLinux. Andrew will briefly describe TurboLinux's commercial closed-source multi-platform (Unix, Linux, NT) clustering solution, EnFuzion, and describe how EnFuzion cluster implementations differ from Beowulf cluster implementations.

For a good introduction to Beowulf, see the Beowulf Web site's Introduction and Overview page listed below.

Web Resources:

  1. The Beowulf Project http://www.beowulf.org
  2. Introduction to Beowulf http://www.beowulf.org/intro.html
  3. Beowulf 2nd Generation Clustering http://www.scyld.com/
  4. Beowulf Underground http://www.beowulf-underground.org/
  5. Top Clusters http://www.topclusters.org/
  6. IEEE Computer Society/Clusters http://www.ieeetfcc.org/
  7. Jacek's Beowulf Page http://www.sci.usq.edu.au/staff/jacek/beowulf/
  8. Beowulf FAQ http://www.dnaco.net/~kragen/beowulf-faq.txt
  9. TurboLinux http://www.turbolinux.com
  10. TurboLinux EnFuzion http://www.turbolinux.com/products/enf/index.html
  11. BSP / Bulk Synch. Parallel Model http://www.bsp-worldwide.org

Book:

Building Linux Clusters by David HM Spector, O'Reilly, 2000

Giveaways:

Caldera has donated three "cartons" of Caldera OpenLinux 2.3 for giveaways at Unigroup meetings. These are full, boxed distributions (3 CDs + Manuals). OpenLinux is a leading commercially supported Linux Operating System! We gave out most of them at our January meeting, but we still have some left to give out (raffle off) as door prizes at this meeting.

Andrew Flesch of TurboLinux should be bringing a variety of giveaways to this meeting which should include: White Papers, T-Shirts and CD-ROMs.

Unigroup appreciates Caldera's and TurboLinux's support!


Description of Talk:

Few of us are aware of how much our everyday life depends on supercomputers. The price of an ordinary good, such as a pair of jeans, is carefully calculated using complicated market simulations. The weather map we see on TV is a result of another high performance parallel application. There is an obvious need for supercomputing power in the academia. The money, the lack of robust funding for a commercial MPP (Massively Parallel Processors) resp., in the later case creates a need for computing technologies that a Beowulf cluster is an example of.

Beowulf clusters fall somewhere between MPP and NOW (Network of workstations). The nodes in the cluster are dedicated to the cluster and they are connected via their own private network. All the nodes are within the administrative jurisdiction of the cluster. The building of such cluster is based on the idea of providing COTS (Commodity off the shelf) to satisfy given computational needs. The similar philosophy applies to the software too, the OS (Linux) and the tools are open source, public and therefore easy to get. That is pretty much the definition of a Beowulf cluster. Surprisingly, there is no "Beowulf" software to be installed. Beowulf is a technology of clustering Linux Computers to form a parallel, virtual supercomputer. [Note that Scyld's Beowulf v2 implementation does boot a special cluster operating system on remote nodes, Unigroup is planning a followup meeting on this technology].

Of course, the message passing libraries like MPI and PVM are used in order to increase the productivity. The parallelization of a given application is achieved by first identifying the concurrent parts of the application and then by executing them on different processors with the help of the message passing libraries. The way the application is broken into the parallel parts is perhaps both the most important and the most difficult (at least if we seek to get the maximum performance from a given hardware).

There is no prescribed way how to build "genuine" Beowulf cluster. Jozef will present the setup he used in the past as an illustration. Also, he will illustrate the parallelization on one of his research problems.


Speaker Biography:

Jozef Skvarcek studied in the Ph.D. program in Physics at the City University of New York between 1994 and 2000, specializing in theoretical quantum optics.

While working as a research assistant in the group of Prof. M. Hillery and Prof. J. Bergou at Hunter College, he created and ran high performance parallel programs that simulated physical processes, namely the interaction between an atom and electromagnetic field inside a microwave cavity. Jozef designed and implemented a Beowulf cluster to perform those simulations.

After graduation in 2000, Jozef started to work at Datek Online. Jozef teaches undergraduate classes in Astronomy at Hunter College and he continues to stay in touch with his former research group.


Complimentary Refreshments and Food will be served.
This now includes salads and sandwiches (eg. turkey, roast beef, chicken, tuna, grilled eggplant)!

Please join us for this meeting, you won't want to miss it!

[Unigroup Home]

unilist@unigroup.org