wescott@LNIC1.HPRC.UH.EDU (Andrew M. Wescott) (08/15/89)
I have been playing with the dn10000 and several other parallel machines for awhile, and I have a nagging suggestion that I would like to make to those in Chelmsford involved with the Prism O/S and compilers. So here goes... We bought dsee.v.3.3.2.p for one reason: concurrent builds on a multiple cpu dn10000. I don't care about cross-compiled concurrent builds with our dn3500s that are coordinated with the dn10000s builds. I want to spawn concurrent builds across a multiple cpu dn10000. I have read all the dsee documentation, and you certainly do concurrent builds by replicating the node name of the dn10000 several times. I have to confess that dsee is an impressive piece of software, but its simply too sophisticated for most of our users. We have users that write large simulation codes that run on many parallel/vector UNIX or UNIX-like platforms. They don't want to be bothered with learning something like dsee on one platform. Okay, so what am I really asking for? Why not have a parallel UNIX make utility for multiple cpu dn10000s? As an example, consider the DYNIX O/S that Sequent runs on the Symmetry and other machines. The enhancements to Sequent's make include a " make -P [number of compile processes] " . If you have 20 cpus, and say 20 independent compiles to do within a makefile, then make -P20 will spawn off 20 compile processes. Now on a dn10000, you would probably only do make -P6 at most, but I think you can see my point. As an alternative to "make -P", you can set the environment variable PARALLEL to whatever value -p[] would have on the command line. Both the dn10000 and dsee are excellent products, but this just seems like the natural thing to do at the O/S level on a multiple cpu machine. Sincerely, Andrew M. Wescott University of Houston Department of Chemical Engineering