Date: Tue, 11 May 1999 13:27:41 +0200 From: "Samer, Michael, IN" <Michael.Samer@Ingolstadt.BERTRANDT.com> To: 'Guy Helmer' <ghelmer@scl.ameslab.gov> Cc: "FreeBSD Questions (E-Mail)" <freebsd-questions@FreeBSD.ORG> Subject: AW: Cluster?! Message-ID: <DE7D44483D7ED211BF3300A0C93B227748B561@in_sv_off>
next in thread | raw e-mail | index | archive | help
Hey, our app consists of ~100 splitts of matrixes with 1mil*1mil fields. A = lot of fields depend on others and most time is used for calculation (a 250MHz = SGI Octane R12000 appr. 1-3weeks) and there is a SMP Version of PamCrash available, so I guess it's scaleable. I've an ALR Evolution 6x with = 6*PPro's at my home, but it's depends on the efficency of the program = (Photoshop) to run good or not. And it doesn't have an eficency more than 40% than one (same calc. than your's). It seems that a cluster would be of better = (and cheaper!) value, than MultiP machines. But I guess the OS is very significant for the performance.=20 78% is really great! Invites to do a rendering of highest quality and = 2048 resolution for a next Starwars movie Part 1 1/2....=20 I have already thought about gigbit Clustering (switch! with 32GBit Backplane). Driver are still a problem, but for that money it would = surely be possibe to let a drive be created... or so?! But great to have a = card already natively supported. Baynetwork/NetGear is not broadly known in Germany and so difficult to get. To get a Foundry Netw. Hub was also = not so easily , but surely possibly. I guess I will try an F-Ether Switch = first and then a gigbit Switch for testing. We will see! Thanks for your hint/test! Greetings Sam > -----Urspr=FCngliche Nachricht----- > Von: Guy Helmer [SMTP:ghelmer@scl.ameslab.gov] > Gesendet am: Dienstag, 11. Mai 1999 04:02 > An: rick hamell > Cc: Samer, Michael, IN; FreeBSD Questions (E-Mail) > Betreff: Re: Cluster?! >=20 > On Mon, 10 May 1999, rick hamell wrote: >=20 > > > Does anyone have experience (on universities or research) with = Cluster > and > > > the efficency?=20 >=20 > Efficiency depends on the distributed application :-) If your app = doesn't > require extremely low latency or extremely high bandwidth for its > interprocessor communication, then it ought to run well. A sample = app > that I use for benchmarking runs at 78% efficiency (e.g., 0.78 * 64 * = the > speed of the app on a single PPro 200) on 64 PPro-200 processors = connected > with Fast Ethernet (although the PC's are running Linux). Some apps = run > very well on clusters, and other apps run poorly. >=20 > > I don't know how much Alphas cost versus regular PCs, but you may=20 > > want to take a look at Beowulf. I've heard several reports of = people=20 > > running it under FreeBSD, including an unconfirmed inside sourse at > NASA.=20 > > There are several commercial clustering programs also, though I do = not=20 > > know the name of them off the top of my head. >=20 > A basic cluster is just a group of PC's connected with a fast = network. > Since Bill Paul recently added a driver for the NetGear GE620 ($330!) > Gigabit Ethernet card to FreeBSD, one can build the cluster around a > really fast network for a moderate amount of money. >=20 > > Anyways if you run several fast PC's PII-400s or so with FreeBSD=20 > > in a clusterd environment you should see quite a bit of = performance, and >=20 > > a large cost savings. The best part (IMHO) of using the PCs, you = can > swap=20 > > new ones in and out easily, and be cost-effective at doing so. >=20 > Or, you can buy a cluster of new systems and distribute the old = systems to > be used as desktop PC's :-) >=20 > Guy Helmer, Ph.D. Candidate, Iowa State University Dept. of Computer > Science=20 > Research Assistant, Ames Laboratory --- > ghelmer@scl.ameslab.gov > Research Assistant, Dept. of Computer Science --- > ghelmer@cs.iastate.edu > http://www.cs.iastate.edu/~ghelmer To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?DE7D44483D7ED211BF3300A0C93B227748B561>