Date: Mon, 4 Dec 1995 10:06:22 -0800 From: "Edward A. Lee" To: ealgroup@edison.eecs.berkeley.edu Subject: Mercury proposal Please comment on the following proposal. It would be good to say more about the active messages idea (and probably the other ideas). If you send me paragraphs, I will incorporate them. Thx. Edward Proposal for Equipment Donation Submitted to Mercury Computer Systems Principal Investigator: Edward A. Lee Pursuant to our meeting on Nov. 15 with Gary Olin, Barry Isenstein, Arlan Pool, and Michael Krueger, we are proposing a collaborative project in which Mercury will provide us with hardware and we will conduct a set of three research projects. We expect to carry out these projects over a period of two years, beginning as soon as possible. Performance Modeling We will construct in a suitable Ptolemy domain (probably the discrete- event, or DE, domain) the infrastructure required to perform performance modeling of Mercury Raceway configurations. There will be a hierarchy of levels of abstraction supported. In the most abstract, random or deterministic execution times will be provided by the designer and the hardware configuration will be modeled at the level of packet exchanges. We will develop visualization tools to help evaluate a design. A less abstract performance modeling tool will begin with a dataflow graph specification of an application. It will measure the execution times of the actors in the graph using first the host workstation and second the target processor on our Mercury prototype. Using these measurements, it will partition the graph for parallel execution on any number of processors with any Raceway interconnect (using adaptations of existing Ptolemy schedulers). It will then combine the schedules and the execution time measurements to construct a more detailed performance model. An interative refinement strategy will be supported where the Raceway interconnect, the number of processors, or the dataflow graph can be modified. Rapid Prototyping We will interface Ptolemy to the Talaris system and define a Ptolemy target capable of generating code for our hardware configuration. Again, the application will be specified as a dataflow graph. The target will manage downloading the code, running the application, and monitoring performance and computational results. We will design this target to support a wide variety of Mercury configurations. Of course we will only be able to fully validate it on configurations supported by our hardware. Active Messages Layer We will investigate the role of an active message layer of abstraction over the raw Raceway packet handling. Active messages are a low-overhead abstract network interface that are currently supported on TCP/IP networks and Myrnet networks. Supporting them on Raceway networks would insulate much of the software infrastructure from details of the Raceway implementation. If it proves to be appropriate, we will implement the active message abstraction on our Raceway configuration. Requested Equipment We would need a fully configured system using a Sparc/Solaris host with an ethernet interface and card cage. The processors can be i860s to start with, although we would prefer to work with PowerPC and Shark processors. We are particularly interested in working with the combination of PowerPC and Shark in a heterogeneous configuration, although it makes sense to start with an all PowerPC homogeneous configuration. We would need the Racetrack kit, including the Xilinx interface, and the Talaris software. Deliverables All software resulting from this project will be distributed under the standard Ptolemy license agreement bearing a university copyright. This is a very liberal agreement that permits open-loop commercialization. The complete text of the agreement is given below as an appendix. In addition, we expect to publish papers that describe the results of the research projects. ------------------------------------------------------------------ Edward A. Lee Professor phone: 510-642-0455 EECS Dept., Cory Hall fax: 510-642-2739 University of California email:eal@eecs.Berkeley.EDU Berkeley, CA 94720 http://www.eecs.berkeley.edu/~eal APPENDIX: the current copyright agreement for the Ptolemy system. Copyright (c) 1990-1995 The Regents of the University of California. All rights reserved. Permission is hereby granted, without written agreement and without license or royalty fees, to use, copy, modify, and distribute this software and its documentation for any purpose, provided that the above copyright notice and the following two paragraphs appear in all copies of this software. IN NO EVENT SHALL THE UNIVERSITY OF CALIFORNIA BE LIABLE TO ANY PARTY FOR DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF THE UNIVERSITY OF CALIFORNIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. THE UNIVERSITY OF CALIFORNIA SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS ON AN "AS IS" BASIS, AND THE UNIVERSITY OF CALIFORNIA HAS NO OBLIGATION TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.