From schaffer@helix.nih.gov Mon Dec 2 17:07:27 1996 Return-Path: Received: from ray.nlm.nih.gov by frodo.nlm.nih.gov id RAA00949; Mon, 2 Dec 1996 17:07:25 -0500 Received: from helix.nih.gov by ray.nlm.nih.gov id RAA07583; Mon, 2 Dec 1996 17:05:22 -0500 Received: (from schaffer@localhost) by helix.nih.gov (8.8.3/8.8.3) id RAA27532 for ncbi-seminar@ray.nlm.nih.gov; Mon, 2 Dec 1996 17:05:21 -0500 (EST) Date: Mon, 2 Dec 1996 17:05:21 -0500 (EST) From: Alejandro Schaffer Message-Id: <199612022205.RAA27532@helix.nih.gov> To: ncbi-seminar@ray.nlm.nih.gov Subject: Parallel Computing/Genetics Seminar Dec. 20 Content-Length: 2002 X-IMAPbase: 1000759407 1 Status: RO X-Status: X-Keywords: X-UID: 1 Anyone who wishes to meet with Prof. Zwaenepoel individually should send me e-mail at schaffer@ray.nlm.nih.gov. ---------------------------------------------------------------------- TreadMarks: Shared Memory Computing on Your IBM SP2 Willy Zwaenepoel Department of Computer Science Rice University December 20, 1996 NIH Bldg. 38A, 8th floor conference room 11AM The IBM SP2 is one of the most successful "cluster machines", present at many laboratories including NIH. The common parallel programming model on the SP2 and on most cluster machines is message passing, exemplified by packages such as PVM or MPI. Unfortunately, many scientists have found message passing hard to use in comparison to the shared memory model available on SMPs like the SGI Power Challenge. TreadMarks is a software package that provides a shared memory programming environment on an IBM SP2. The package is also available on other cluster architectures and on networks of workstations. Most common Unix platforms and many different networks are supported. A parallel program written for TreadMarks therefore enjoys a great measure of portability, spanning clusters, networks of workstations, and SMPs. While the appeal of shared memory programming has been well known for some time, early software implementations have suffered from poor performance. I will explain the nature of these performance problems, and discuss the techniques that TreadMarks uses to overcome them. I will demonstrate the programmability and efficiency of TreadMarks by discussing the parallelization of FASTLINK, a popular package for genetic linkage analysis, the computational step in disease gene location. Parallel FASTLINK has recently been used in several genetic linkage studies at NIH, including the discovery of a linkage for Parkinson's disease on chromosome 4. Hosts: Alejandro Schaffer(schaffer@helix.nih.gov) and Jim Tomlin (jtomlin@helix.nih.gov)