U.C. Berkeley Millennium: System Architecture and Engineering

Progess Reports

1998 Q1 An initial version of the Solaris x86 software distribution was developed.

AM II was ported to SolarisX86 on the Pentium. This required extensive development and testing to resolve endianess and driver issues.

Detailed analysis was conducted of synchronization operations on the PPro to support the memory-based Active Messages protocol.

Numerous benchmarking efforts were conducted. This includes benchmarks of the C++ based finite element models use by the Civil Engineering group.

Glunix was ported to Solaris x86

A number of tools were ported to the Millennium environment and a tutorial was established.

A VIA port was started.

Root Solaris and NT domain servers were established.

1998 Q2 The major progress this quarter was that we received a 1.7 Million $ grant from the NSF CISE Research Infrastructure program to complete the Millennium design and to investigate computational enconomies as a basis for large scale distributed systems, with interactive visualization.

We completed the port and integration of a complete NOW software environment on the quad-PPro Millennium cluster, including multiprotocol Active Messages, Glunix, and MPI. The most widely used graph partitioner, PETc was ported on top of our MPI. The Titanium compiler came up.

A set of tutorials was written to facilitate use. These were tested by several outside users.

A VIA implementation was built and performance analysis was conducted against numerous research user-level networks. This work is to be published in SC98.

The study of alternative cluster area networking technologies was conducted and the Synfinity I network from Fujitsu System Technology Lab was received for testing. ServerNet from Compaq is in progress. Initial evaluation of gigabit ethernet was performed.

An initial programming paradigm for construction of persistent, customizable services, called I-Space, was developed in Java. This can be located at

Development of the basic infrastructure for a computational economy progressed substantially.


February 1999