From: Randy Crawford Newsgroups: comp.parallel Subject: Re: Newbie questions Date: 20 Dec 1998 06:44:48 GMT Organization: ISPNews http://ispnews.com Approved: bigrigg@cs.cmu.edu Message-Id: <75i6d0$hj1$1@encore.ece.cmu.edu> Originator: bigrigg@ece.cmu.edu Duane K. Dougal wrote: [...] > I have this wild (?) idea of taking a bunch (3 or 4) of > readily-available surplus Pentium PCs and somehow connecting them > together to achieve a parallel processor not unlike the machine > described above. My hope is to create from obsolete technology a > parallel machine that, transparently, will meet or exceed the latest > single-processor PC technology in speed and performance. I'm not > interested in a network, but in one machine using multiple processors. Then you'll need a custom interconnect and a fast protocol and custom routers to move data among the motherboards. That *will* be wild. > My questions are these: > * Is this a stupid idea or is it actually feasible? (Disregarding strictly hardware-based solutions... Too hard.) Feasible, but inherently self-limited due to the high latencies in message passing across a slowly-switched interconnect (usu. ethernet). Some kinds of highly localized codes will run fast, but if there's any memory sharing/movement, it won't speed up as you run the same job with more CPUs. (Too much time is spent in communicating among the CPUs). > * If it is feasible, can anyone guide me to information on how to > accomplish such a project? Search on Beowulf (a solution arising at NASA Goddard that uses Linux) or visit http://www.beowulf.org/ . There are several NT solutions out there. Look at Condor ( http://access.ncsa.uiuc.edu/Archive/AOarchive/Condor.html ), or Symera ( http://access.ncsa.uiuc.edu/CoverStories/Symbio/Symbio.html ) or NCSA's NT cluster ( http://access.ncsa.uiuc.edu/News/980804.LSFsupercluster.html ) for a custom NT 256 Pentium solution they're exploring). > * Can I use something like NT for an OS, or do I have to use a UNIX > variant (e.g., Linux, et al.)? See the above. Also look into Globus ( http://www.globus.org/ ), or Legion ( http://legion.virginia.edu/ ). > * Are there any commercially-available or stable (production quality) > research-grade products to allow me to make this a reality? Sure, but the real challenge is hardware, not software. > * Would such a project be cost prohibitive (the cost of making it work > outweighing the derived benefits)? -- Randy Crawford crawford@net66.com rlc@urbana.sgi.com N=1 ==> P=NP -- Articles to bigrigg+parallel@cs.cmu.edu (Admin: bigrigg@cs.cmu.edu) Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel