[Beowulf] Newbie question.
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Ryan Sweet rsweet at aoes.comFri Oct 29 02:18:59 PDT 2004
- Previous message: [Beowulf] Newbie question.
- Next message: [Beowulf] Updated NCBI rpms released for the 2.2.10 toolkit
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On Thu, 28 Oct 2004, Currit, Dennis wrote: > We are thinking of putting up a cluster to run MSC Nastran and have about > $25,000 budgeted for hardware. Have you also discussed licensing with MSC? It may have changed recently, as my experience here is a few months old, but Distributed Parallel features are licensed differently than the SMP Parallel features, and are significantly more expensive. > Is this enough to get get started? Certainly it is enough to buy a few nodes, but also if you are just getting started with clustering it may be good to think a couple of times about the direction you want to go and to analyse the associated costs of things like additional power consumption, air-conditioning, etc.... If you already have a sufficient server room infrastructure (racks, power, ac) then perhaps this isn't much of a consideration. > Any suggestions as to what I should buy? We are currently running NASTRAN with success on opteron, though we have also had good experience with PIV and Athlon MPs. I think the best thing to do is to try and benchmark your actual jobs on a few different systems if you can. > Currently we are running large jobs on > an older AIX multiprocessor system and smaller jobs on a Xeon 2.8 system, so > I think even a small cluster should be an improvement. It depends upon the jobs that you are running. If your jobs do benefit well from running in SMP on the AIX system, then you may also have good efficiency with DMP on a cluster. We have had mixed results with our structural analyses: some had near linear speedups on smallish (<16 cpus) cluster runs, and others would gain only about 20% for each cpu added. For many of our NASTRAN runs high speed disk IO is nearly as important as as the CPU. Find out what your job mix is like, and then spend your money accordingly (you may want to splash on fast SATA local disks for scratch space, or on more RAM to use RAM disks for scratch). Also have the engineers read through the manual regarding the DMP options because you have to pay a bit more attention to how your jobs are configured when using distributed parallel. good luck, -Ryan -- Ryan Sweet <ryan.sweet at aoes.com> Advanced Operations and Engineering Services AOES Group BV http://www.aoes.com Phone +31(0)71 5795521 Fax +31(0)71572 1277
- Previous message: [Beowulf] Newbie question.
- Next message: [Beowulf] Updated NCBI rpms released for the 2.2.10 toolkit
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
