Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!ncrlnk!ncrcae!hubcap!wen-king From: wen-king@csvax.caltech.edu (Wen-King Su) Newsgroups: comp.parallel Subject: Re: using network machines as compute servers Message-ID: <7345@hubcap.clemson.edu> Date: 6 Dec 89 14:00:44 GMT Sender: fpst@hubcap.clemson.edu Lines: 64 Approved: parallel@hubcap.clemson.edu >From: carriero-nicholas@YALE.EDU (Nicholas Carriero) wen-king@csvax.caltech.edu (Wen-King Su) writes: >>>From: larsa@nada.kth.se (Lars Andersson) <>>>picked up by the "slaves" as each completes its current task. When a slave <>>>picked up and stored by the master (depending on the problem, one might want <>> <>This can't be that difficult. You write a batch process (listed below). > <[54 Lines of code to implement a batch process deleted.] > think about (let alone design and implement) a 'batch' process she worker (i.e. the control structure for the whole problem), but this be given in any number of systems. [a bunch of C-Linda lines deleted] First of all, you are right; the same solution given in C-Linda can just as easily be written in the generic C language used by Cosmic Environment. Let me make it clear that CE is not produced by somebody sitting around, thinking about how one might program a multicomputer. We do not have a person who earns his degree by creating the Cosmic Environment. We are all multicomputer users and application programmers --- I am an author of CE and my project is distributed discrete-event simulation. Cosmic Environment is the result of people like me who wanted to protect our investments. All too frequently, we find ourselves wasting valuable time supporting the same programs on several incompatible systems. We find that our programs got stuck with systems that becomes obsolete. We also find that, although many systems are full-featured and come with everything plus a kictchen sink, they are almost always mis-focused, and we are unable to do want we wanted without a lot of work. We understand that the programmer knows exactly what he need, and the system aught to give him all the tools he will need, no less but no more. (We mean it, and to illustrate that, the programming manual for CE is only 24 pages long, typesetted in 11 points fonts.) We have settled on a set of simple, easy to understand, and easy to implement C primitives which we share with the public in the form of the Cosmic Environment. It makes our jobs simpler and it protects the value of our work. We use it daily for applications on several multicomputers we own and for the set of SUN workstations we have. This week, I am using a CE program that runs a batch which serves a set of programs running on a dozen workstations. Each program runs a copy of the CMU COSMOS circuit simulator to evaulate the performance of one student's circuit layout for the final grade in their VLSI design class. Here, my requirement is a distributed system that allows me to run a program that I do not want to modify (or know how to modify). A batch in CE comes handy. Now, back to the problem at hand. If the programmer finds it suits his need to approach a problem using the master/worker model you described, then no problems. If you or I want to suggest an alternative solution, no problems either. But if the programmer needed a batch, I know I would want to have simple ways to build my own d*mn batch. /*------------------------------------------------------------------------*\ | Wen-King Su wen-king@vlsi.caltech.edu Caltech Corp of Cosmic Engineers | \*------------------------------------------------------------------------*/