FM (Fast Messages) Library for the Cray T3D

vijayk@orange-utan.cs.uiuc.edu (Vijay Karamcheti)
Thu, 9 Feb 1995 16:26:01 GMT

          From comp.compilers

Related articles
FM (Fast Messages) Library for the Cray T3D vijayk@orange-utan.cs.uiuc.edu (1995-02-09)
| List of all articles for this month |

Newsgroups: comp.arch,comp.compilers,comp.parallel.pvm
From: vijayk@orange-utan.cs.uiuc.edu (Vijay Karamcheti)
Summary: fast messaging primitives for the Cray T3D
Keywords: parallel, available, WWW
Organization: University of Illinois at Urbana
Date: Thu, 9 Feb 1995 16:26:01 GMT

FM - Fast Messaging on the Cray T3D
-----------------------------------


    The FM library contains fast messaging primitives which exploit
special features of the Cray T3D hardware to provide very low latency
for short messages. FM provides an order of magnitude lower latency
than Cray's PVM and achieves performance comparable to SHMEM get while
providing a message-passing interface.


    The FM library provides two distinct sets of primitives which make
use of the T3D fetch-and-increment and atomic swap hardware
respectively. The fetch-and-increment primitives are optimized for
the lowest possible latency and are suitable for situations with light
communication traffic. The atomic swap primitives eliminate output
contention at the cost of slightly higher latency, but by doing so can
deliver robust performance even for heavy and unbalanced traffic loads.




    Release 1.0 of the library is now available from our WWW server:


        http://www-csag.cs.uiuc.edu/projecs/communication/t3d-fm.html


    The library can also be accessed on the T3D at Pittsburgh
Supercomputing Center (mario.psc.edu) from the directory:


        /usr/users/9/karamche/FM-1.0


    The release contains the source files (C and Assembly), the library
(libFM.a), and an include file which provides the function prototypes.
The release directory also contains the usage manual and a copy of a
paper analyzing the performance of the two sets of FM primitives. The
latter is a preliminary version of the paper which will appear in the
Proceedings of the 22nd International Symposium on Computer
Architecture (ISCA'95).




    Please contact me if you have any questions, comments or problems.


Vijay Karamcheti
vijayk@cs.uiuc.edu
(217) 244-7116


Concurrent Systems Architecture Group
Department of Computer Science
University of Illinois at Urbana-Champaign
1304 W. Springfield Avenue
Urbana, IL 61801
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.