shir@luck.ece.orst.edu (Shirish S. Pargaonkar) (04/26/91)
Hi,
I came across a paper in 1989 IEEE conference on parallel processing
by Kai Li and Richard Schaefer which is about implementing shared virtual
memory on iPSC/2 hypercube.
I am exploring the idea of such implementation on NCUBE/7 hypercube
at our university.
So far I have two papers. First one is above mentioned and the second
one is IVY: A shared virtual memory system for parallel processing" by
Kai Li. I am in the process of getting PhD thesis of Kai Li who proposed
this idea.
I would appreciate any suggestions or ideas you folks may have that
you want to share with me. I would like to know if it is possible to
implement shared virtual memory on NCUBE hypercube. Since NCUBE does not
have hardware MMU is is possible to simulate it... something on this line. Or if
NCUBE has already implemented it.
Thanks in advance.
--
=========================== MODERATOR ==============================
Steve Stevenson {steve,fpst}@hubcap.clemson.edu
Department of Computer Science, comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabellmikek@ios.Convergent.COM (Michael Koster) (05/01/91)
shir@luck.ece.orst.edu (Shirish S. Pargaonkar) writes: > I came across a paper in 1989 IEEE conference on parallel processing >by Kai Li and Richard Schaefer which is about implementing shared virtual >memory on iPSC/2 hypercube. > So far I have two papers. First one is above mentioned and the second >one is IVY: A shared virtual memory system for parallel processing" by >Kai Li. I am in the process of getting PhD thesis of Kai Li who proposed >this idea. I have been interested in this type of memory architecture for some time now, and I am in the process of organizing a bibliography of papers which deal with the subject of shared virtual memory. My focus is somewhat more oriented to hardware support/enhancement, so maybe some of this collection will help with your MMU problem... >implement shared virtual memory on NCUBE hypercube. Since NCUBE does not >have hardware MMU is is possible to simulate it... something on this line. I'm not intimately familiar with hardware memory management features of NCUBE, but if it has no MMU whatsoever, i.e. no virtual memory, then it must be difficult to do anything ;-). I think memory management for SVM systems requires dynamic address mapping from program space to local memory(cache) space to be able to migrate data on demand (this is the primary feature) while effectively utilizing the physical memory at each node. The lack of a hardware MMU would seem to make this very difficult. I don't know if your scope includes hardware... I am not aware of any SVM or SVM-like system implemented on the NCUBE. When I complete the bibliography list, I can mail it, or post if there's some interest... Michael. -- Michael Koster, mikek@ios.Convergent.COM Newsgroups: comp.parallel Approved: parallel@hubcap.clemson.edu Date: Tue, 30 Apr 91 11:07:10 +0200 From: caerts@esat.kuleuven.ac.be Subject: Bonn PCA SIG RTIS Message-ID: <1991Apr30.110839.5173@esat.kuleuven.ac.be> Date: 30 Apr 91 11:08:39 GMT Organization: Katholieke Universiteit Leuven,ESAT - dep. Elect. Engineering, Belgium News-Moderator: Approval required for posting to comp.parallel Lines: 18 Bonn PCA Applications Group SIG Real-TIme and Industrial Systems Dear PCA members, All received progress reports have been accepted and will be scheduled in the Applications Group Session (monday or wednesday). Because there is considerate common interest for some reports of the SIG on Signal Processing and Neural Networks, we will join them for some presentations. Detailed schedule will be available very soon. See you in Bonn Chris Caerts email : caerts@esat.kuleuven.ac.be -- =========================== MODERATOR ============================== Steve Stevenson {steve,fpst}@hubcap.clemson.edu Department of Computer Science, comp.parallel Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell