haines@handel.cs.colostate.edu (Matt Haines) (02/12/90)
I just finished an article on Shared Virtual Memory Systems by Kai Li
and Paul Hudak (ACM TOCS Vol. 7, No. 4, November 89).
I am interested in knowing if anyone else has proposed
such a system for implementing a shared virtual address space on a
distributed memory machine. I am interested in investigating this
subject, and I'm wondering if there are any other approaches that have been
proposed. I would appreciate any references on this subject.
Matt.
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
"Don't take life so seriously - | Matt Haines
You'll never get out of it alive!" | Colorado State University
| haines@CS.ColoState.Edu
Thierry.Priol@irisa.fr (Thierry Priol,TB131,Equipe API,9936200-547,) (02/13/90)
>From article <7955@hubcap.clemson.edu>, by haines@handel.cs.colostate.edu (Matt Haines): > I just finished an article on Shared Virtual Memory Systems by Kai Li > and Paul Hudak (ACM TOCS Vol. 7, No. 4, November 89). > I am interested in knowing if anyone else has proposed > such a system for implementing a shared virtual address space on a > distributed memory machine. I am interested in investigating this > subject, and I'm wondering if there are any other approaches that have been > proposed. I would appreciate any references on this subject. > > Matt. > -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- > "Don't take life so seriously - | Matt Haines > You'll never get out of it alive!" | Colorado State University > | haines@CS.ColoState.Edu Kai Li & R. Schaefer have published another article which present a Shared Virtual Address Space in The last Conference (1989' International Conference on Parallel Processing I-125 to I-132). There are also some papers on this subject in the conference. We work on this subject at IRISA and we begin to implement such virtual memory on a hypercube iPSC/2. We have an application (ray-tracing) which uses such shared-memory. Using a shared-memory programming model for parallelizing the ray-tracing algorithm on a distributed memory parallel computers seems to be the best approach on a DMPC. We have a paper which compares two parallel algorithms, one with a message-based programming model and the other with a shared-memory programming model. However the last uses a shared-memory services implemented inside the user-application. A kernel implementation is better and it is our goal now. Thierry. Thierry PRIOL Phone: 99 36 20 00 IRISA / INRIA U.R. Rennes Fax: 99 38 38 32 Campus Universitaire de Beaulieu Telex: UNIRISA 950 473F 35042 RENNES CEDEX - FRANCE E-mail: priol@irisa.fr
jhs@icase.edu (Joel Saltz) (02/26/90)
We have a spectrum of work going on in this area at ICASE, we are both designing distributed compilers (Kali Fortran) and designing tools to be used directly by programmers (PARTI). All of this work is designed to provide the illusion of shared memory on distributed machines, at least when using parallel loop constructs (of course efficiency is a whole other story ... but if data is well mapped it sometimes works out well). Our favorite applications areas are in unstructured meshes, sparse matrices and adaptive problems although the compiler effort also addresses regular problems. Here are a few references: (we can supply preprints) Koelbel C. and Mehrotra P. and Van Rosendale J. Supporting shared data structures in distributed memory architectures, To appear in PPoPP March 1990 J. Saltz and K. Crowley and R. Mirchandaney and Harry Berryman Run-time Scheduling and Execution of Loops on Message Passing Machines, to appear in Journal Parallel and Distributed Computing, April 1990 S. Hiranandani and J. Saltz and H. Berryman and P. Mehrotra Compiler Directed Data Migration on Distributed Memory Machines In Fifth Distributed Memory Computing Conference, Charleston, SC, April 1990. Joel Saltz ICASE NASA Langley Research Ctr Hampton VA 23665