Third Java Grande Forum Meeting Aug 6-7 1998

http://www.javagrande.org (as of mid September)

http://www.npac.syr.edu/javagrande/charterindex.html (before mid September)

http://www.jhpc.org/grande/

http://math.nist.gov/javanumerics/

The Third Java Grande Forum meeting was held Aug 6-7 98 in Palo Alto. It was sponsored by Sun Microsystems (Siamak Hassanzadeh), and coordinated by Geoffrey Fox with George Thiruvathukal as secretary. Grande applications are large-scale applications typical of HPCC, scientific and engineering computations, or distributed simulations. The goal of the forum is to further community activities that will make Java a much better (and probably the best) programming environment for Grande applications. The first two meetings of the Forum were March 1,98 and May 9-10,98(covered in an earlier trip report). This meeting had over 30 participants from academia, industry and government.

The next meeting will be public presentations and debate with a 3 hour session (http://www.npac.syr.edu/javagrande/SC98JGpanels.html) at SC98 on the morning of Friday November 13.

The first meeting (http://www.npac.syr.edu/javagrande/JavaGrandeForum.html) established the concept while the second meeting established the issues and produced good interim reports.( http://www.npac.syr.edu/javagrande/javagrande2.html) . The third meeting was also successful and refined the reports -- setting up the process to present our recommendations to both Sun and the Grande community.

The meeting had interesting plenary presentations on a variety of topics. Jini(http://java.sun.com/products/jini/index.html ) offers a general approach to distributed resource registration and discovery and seemed applicable to both hardware and software Grande components. Note that this base technology has a Linda-like distributed computing environment Java Spaces (http://java.sun.com/products/javaspaces/index.html) built (conceptually if not in practice) on top of it. Henry Sowizral (who in times gone by pioneered the Time Warp approach to event driven simulation) described the extensive (but focussed) Java matrix capability in Java3D graphics framework. It is remarkable how many classes 3 by 3 and 4 by 4 matrices can produce! This was contrasted to the scientific matrix package Jama (http://math.nist.gov/javanumerics/jama/ ) developed by NIST and MathWorks, which was announced at our meeting in Cleve Moler's presentation. There was a lively presentation from Professor William Kahan (UC Berkeley) and Joseph D. Darcy on "How Java's Floating-Point Hurts Everyone Everywhere" (http://www.npac.syr.edu/javagrande/JAVAhurt.pdf ) and why Sun's proposed floating point changes were flawed (http://www.npac.syr.edu/javagrande/jgrandefromucb.pdf ).

Tim Wilkinson described his company (http://www.transvirtual.com/) with its open Java VM Kaffe available freely. He also discussed optimization issues and noted that obviously he was not content to meet C++ performance but aimed at raw C and Fortran levels. Marvin Solomon from the Wisconsin Condor group described their distributed computing system and how it can both use Java as a development tool and support Java Grande applications.

We continue to have two major working groups with a crosscutting interest in benchmarks. In this respect note the new Java benchmark collection from NIST at http://math.nist.gov/scimark/ .

  1. Numerics and Libraries led by Ron Boisvert and Roldan Pozo of NIST.
  2. Applications and Parallel/Distributed Computing led by Dennis Gannon from Indiana.

The numerics working group (http://www.npac.syr.edu/javagrande/jg3numsumm.html) reviewed the interim report and affirmed their basic positions on the issues of complex, efficient classes, operator overloading, and multidimensional arrays. However the section on floating point was actively discussed and this part needed substantial revision and further a response will be generated to Sun's recent proposal on Java floating point.

The second working group carefully reviewed issues and decided that their request to Sun should only address the issue of RMI performance where capabilities to add faster transport layers are needed. Areas such as the scaling and performance of the JavaVM needed further study. We also discussed "Seamless computing" and started a working group to study systems such as UNICORE, WebSubmit, Condor, Globus and Legion to extract the features of a "Java framework for Grande Computing". We also agreed to discuss the MPI Java binding while the collection of a set of "application" benchmarks was agreed.