From chemistry-request -x- at -x- ccl.net Mon Nov 3 04:04:28 2003 Received: from ozone.cs.vu.nl (ozone.cs.vu.nl [130.37.24.158]) by server.ccl.net (8.12.8/8.12.8) with ESMTP id hA393thP032053 for ; Mon, 3 Nov 2003 04:03:56 -0500 Received: from chem.vu.nl (far27.chem.vu.nl [130.37.148.106]) by ozone.cs.vu.nl with esmtp (Smail #87) id m1AGacn-000NHbC; Mon, 3 Nov 2003 10:03 +0100 Message-ID: <3FA6047B.4070200(at)chem.vu.nl> Date: Mon, 03 Nov 2003 08:32:11 +0100 From: Anton Feenstra Organization: Vrije Universiteit Amsterdam - Pharmaceutical Chemistry User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.4.1) Gecko/20031008 X-Accept-Language: en-us, en MIME-Version: 1.0 To: chemistry(at)ccl.net Subject: Re: CCL:Linux Cluster References: <3FA0CEC5.8010502(at)silex.dk> <1510089009.20031031094924(at)bancorp.ru> <3FA2258B.1010809(at)silex.dk> In-Reply-To: <3FA2258B.1010809(at)silex.dk> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-Spam-Status: No, hits=-2.6 required=7.0 tests=EMAIL_ATTRIBUTION,IN_REP_TO,QUOTED_EMAIL_TEXT,REFERENCES, REPLY_WITH_QUOTES,USER_AGENT_MOZILLA_UA,X_ACCEPT_LANG version=2.55 X-Spam-Checker-Version: SpamAssassin 2.55 (1.174.2.19-2003-05-19-exp) Kenneth Geisshirt wrote: > Gregory Shamov wrote: > >> KG> A number of companies build clusters (e.g., the one I work for), >> KG> and you can buy cluster distributions like MandrakeSoft's >> KG> Clustering product. Finally, you can build a cluster by hand (ok >> KG> if the number of nodes is less than 6). >> >> And what happens if the number of nodes becomes 7 ? Or even 8 ? > > > My point is that building a cluster by hand does not scale well. If you > plan a large cluster (32+ nodes) you will need some kind of automated > tool. If you build a 4-node cluster and add a new node (different > hardware e.g., an old office pc) you can do it by hand. Im my experience, if you keep the hardware relatively homogeneous (i.e. at least 16 nodes at a time are identical, and no more than say 2-3 different types of nodes) any number of nodes is as much work as only one. It isn't too difficult to set up a system with net-boot and harddisk images that installs new nodes automatically, and re- installs nodes in case of crashes or other problems. I've done this now for two different 'off-the-shelf' clusters and after initial tweaking of setups (mostly related to network-card and boot-rom specifics) everything is rock-stable and nearly self maintaining... I've never done anything similar with e.g. 'old' office PC's (i.e. in-homogeneous hardware), but it probably is more work than it is worth relative to the average low speed of the 'old' CPU's. -- Groetjes, Anton _____________ _______________________________________________________ | | | | _ _ ___,| K. Anton Feenstra | | / \ / \'| | | Dept. of Pharmacochem. - Vrije Universiteit Amsterdam | |( | )| | | De Boelelaan 1083 - 1081 HV Amsterdam - Netherlands | | \_/ \_/ | | | Tel: +31 20 44 47608 - Fax: +31 20 44 47610 | | | Feenstra(at)chem.vu.nl - www.chem.vu.nl/~feenstra/ | | | "If You See Me Getting High, Knock Me Down" | | | (Red Hot Chili Peppers) | |_____________|_______________________________________________________|