Paul, The GdH implementation is currently used here, and is basically in beta-release form. If you would use it we could wrap up a and X86/Linux version for you. The GUM implementation of GpH and GdH uses very simple communication (mainly point to point, with a single barrier synchronisation, and the communication is restricted to just 3 C modules. As a result it can be readily ported to new communications libraries, e.g. MPI and CMMP (bespoke CM5 library) versions exist.
Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing?
Don't know a lot about Mosix, but if the PVM calls in the 3 GUM communications modules can be replaced with Mosix shared-memory routines that would give a low effort port of GUM to a Mosix cluster. I'd expect Mosix communication to be *much* faster than PVM: it's so generic that it absorbs a lot of time, e.g. probing for a message takes 2000 machine cycles. For small numers of processors a higher-performance implementation could be constructed by adapting the runtime system to effectively use the shared memory, but IIRC Simon Marlowe found there were problems with memory contention with more processors. Phil On Tue, 20 Aug 2002 21:17:28 +0800 paul@theV.net wrote:
Speaking of GpH, I wonder how is GdH coming along? It seems that the installation instruction on http://www.cee.hw.ac.uk/~dsg/gdh/ is still incomplete...
As far as I know, Mosix has nothing to do with PVM, so am I right to say that GpH still needs PVM even you use it on a Mosix cluster? Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing?
Regards, .paul.
On Tue, Aug 20, 2002 at 12:09:28PM +0100, Phil Trinder wrote:
Murray,
There are several parallel Haskell implementations: a survey of them has just appeared in Journal of Func. Prog Vols 4&5 (July & Sept 2002). Implementations are available for o Eden http://www.mathematik.uni-marburg.de/inf/eden/ o GpH http://www.cee.hw.ac.uk/~dsg/gph/
My group works on Glasgow parallel Haskell (GpH) which extends Haskell 98 with a parallel composition combinator. As Simon said the main implemntation, GUM, buys portability using the relatively slow PVM communications library. This doesn't matter so much on distributed memory machines, and we've recently achieved some quite respectable results SunServer shared-memory machines.
Simon Marlowe developed an alternative SMP implementation of GpH a couple of years ago that may be more suitable for a Mosix platform, but I'm not sure of the status of that implementation now.
Phil
On Mon, 19 Aug 2002 18:04:44 +0100 Simon Peyton-Jones <simonpj@microsoft.com> wrote:
Hi Murray
I'm catching up with my email backlog, but I didn't see a reply to your message, so I thought I'd reply. The parallel-Haskell crew are at http://www.cee.hw.ac.uk/~dsg/gph/ There's a mailing list that I'm ccing.
GHC can be built to run on a shared-memory multiprocessor, but we have not built and tested that for some time.
Untrue - both Eden and GpH are currently used on SunServer SMPs.
GPH is aimed more at distributed-memory machines, and has quite a bit more code layered on top of GHC. Details on the GPH page above.
Let me know if I can help. It'd be great to get some more parallel-Haskell stuff going.
Simon
| A group of us at Brooklyn College (City University of New | York) are in the latter stages of setting up a Mosix cluster | that we would like to use for research on, among other | things, parallel execution of Haskell code and run-time | optimization of both automatic and programmed parallelization. | | I would greatly appreciate hearing from anyone who can | provide advice on | an appropriate bibliography on the language implementation and past | work in this area (other than the code in the Haskell compilers and | libraries, of course), approaches, ongoing work, etc. | | Also, I would be most interested in hearing from anyone | interested in joining our effort (no problems with remote | access to the cluster by way of SSH), or making arrangements | for cooperation in research in this area, although, of | course, access to our equipment will be subject to approval | from "higher authorities" (however, note that I really don't | expect difficulty here, since our administration has already | shown considerable interest in supporting the project). | | Replies to the list will be read, but those who wish may mail | me directly at mgross@dorsai.org. | | Thanks in advance for any replies. | | Murray Gross | Adj. Lecturer, Brooklyn College, | City University of New York | | | | _______________________________________________ | Haskell mailing list | Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell |
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS
E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder
_______________________________________________ Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder