From: Xingfu Wu Newsgroups: comp.parallel.pvm Subject: Re: Pros and Cons of PVM, with C and Unix. Date: Wed, 16 Dec 1998 15:29:51 -0600 Organization: Department of Computer Science, LSU Message-Id: <3678264F.8C0891EA@bit.csc.lsu.edu> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit You can find the performance list of PVM in the PVM homepage. I ported PVM 3.3.11 to a special cluster hardware with our communication protocol. We use the protocol to replace TCP/IP. I gave up PVMD, because PVMD is a performance bottleneck. I directly programmed PVM user interface ( included supporting group manage) by our own low level communication subroutines, which are similar to NX subroutines. The loading task method is similar to that of MPI. The bandwidth and latency of user-level applicatons are almost the same as those in the hardware level. They are much better than those on 100MB Ethernet. I think that my porting is very efficient, although I still don't compare the results of other implementations. Using PVM is very easy because PVM is simple, and fewer subroutines than those of MPI. Xingfu Wu "hsh2@cov.ac.uk" wrote: > Hi, > > I am (a newbie) looking for some advantages and disadvanges of using PVM > over > a *real* parallel machine, points specfic to PVM inplementation with C and > Unix would be nice. > > At present I am using PVM for performace evaluation purposes of parallel > applications. I am finding the > results (times) I get flutuate a great deal, the performance of running > applications > on PVM seem to be dependant on many other things e.g. network, other > running processes etc, is this > a disadvantage? can I achieve near constant application performace over a > PVM? > > Of course cost is a big plus point, but what are the > others? > > Regards > > S. > hsh2@cov.ac.uk