Mailing List ArchiveSupport open source code!
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]Re: tlug: linux cluster
- To: tlug@example.com
- Subject: Re: tlug: linux cluster
- From: Karl-Max Wagner <karlmax@example.com>
- Date: Sat, 17 Oct 1998 22:52:34 +0000 (GMT)
- Content-Transfer-Encoding: 7bit
- Content-Type: text/plain; charset=US-ASCII
- In-Reply-To: <3624A00D.BA4B93FA@example.com> from "Andrej Sali" at Oct 14, 98 08:58:53 am
- Reply-To: tlug@example.com
- Sender: owner-tlug@example.com
> Hi, guys, > > Reading your messages, I realized that we are very lucky. > Our computations are embarassingly parallel, so we will > only need to use CLUSTOR of Active Tools to do the following > with each of the nodes, totally independently of each other > (we are already doing this with a few SGI workstations, > including one 20 processor Origin 2000): > > 1) push a couple of files from the root machine to the node > machine (not more than a few megabytes in total) . > 2) do a calculation on the node (from a few minutes to a few hours). > 3) get a couple of files back to the root machine (not more than > 10 MB). > > My guess is that for a 32 processor cluster of Pentium II's, a > 100 Mb/s ethernet "switch" will do the job. So we only need standard > Linux and our program (in Fortran90) on each of the machines; the > installation will be more or less a one time effort (hopefully). > > In comparison to single processor boards, do people find a big decrease > in performance when using two-processor boards, with large > floating-point > intensive programs that rely on large arrays (>> 512kB)? Did you already consider hooking a handful of high end signal processors together and feed and control them from a Linux system ? They are dirt cheap, have extremely powerful interprocessor communications facilities and are rocket fast. The fastest at the moment, the Texas Instruments TMS320C67XX series in fact boasts 1.5 GOPS and something around 500 MFlops with a 100 $ price tag or so. Admitted, the TI DSP architecture is a bit awkward, it is made for extreme speed and efficiency and not for maximum programmer convenience, but with some skillful programming you should be able to extract most of the performance available. Certainly, it requires some hardware building, but it also gives you the best bang for the buck - by far. At least that's the way I'd do things in your place. ================================================================ "It was hell. They knew it. Karl-Max Wagner But they called it karlmax@example.com W-I-N-D-O-Z-E" ================================================================ --------------------------------------------------------------- Next Nomikai: 20 November, 19:30 Tengu TokyoEkiMae 03-3275-3691 Next Meeting: 12 December, 12:30 Tokyo Station Yaesu central gate --------------------------------------------------------------- Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp
- References:
- tlug: linux cluster
- From: Andrej Sali <sali@example.com>
Home | Main Index | Thread Index
- Prev by Date: Re: tlug: sendmail question
- Next by Date: Re: tlug: kanji or romaji for Japanese? (was: parallel-port IDE)
- Prev by thread: Re: tlug: Re: linux cluster
- Next by thread: tlug: More SOHO
- Index(es):
Home Page Mailing List Linux and Japan TLUG Members Links