Talk:Beowulf cluster

Origin of name

The name comes from the legend of Beowulf.

How come? The relationship between a superhuman warrior and distributed computing seems somewhat vague to me. — Preceding unsigned comment added by 216.93.217.7 (talk • contribs) 01:21, 18 February 2005

I wa wondering the same myself. The closest I have found is this beowulf.org. Billlion 15:13, 8 March 2006 (UTC)

It's been six (!) years since I read Beowulf, so my memory is a big hazy. But I believe Beowulf was an incredibly strong warrior who got a Giant's sword. So he could supposedly kill any enemy, if he were 'strong' enough. A distributed computing system, if it were 'big' enough could supposedly solve any problem. (Of course, so too could an ancient, slow computer given enough time. Then again, wind blowing at the enemy would eventually blow away all of his cells). In any case, I doubt there was supposed to be a very exact correlation but if you can't see any connection between a very powerful person and a very powerful computer then you are an idiot. --Carbonrodney 08:59, 27 May 2006 (UTC)

What Is Funny About /. Jokes?

There are constant humorous references to Beowulf Cluster on slashdot.com. Will someone explain what the inside joke is, and what is so funny about it? User:Prophetben 8 August 2005

Nothing is "funny" about it - pathetic, perhaps, but not funny - that died a long time ago. The origin of the "joke" came about shortly after the first Beowulf clusters were announced - thereafter every time a new, more powerful CPU or computer was announced on Slashdot, somebody would say "Imagine a Beowulf cluster of those". At first such statements were valid, if somewhat vacuous - then like so many other trolls it just became Yet Another Stupid Thing To Post. N0YKG 17:19, 15 November 2005 (UTC)

Examples

I removed the two examples of @home projects because they are not Beowulf clusters. They are examples of clusters and distributed computing but not Beowulf. Lotu 21:48, 22 March 2006 (UTC)

Perhaps the Univeristy of Bristol's statistics cluster is an example see http://www.maths.bris.ac.uk/admin/computer/scone It's a 24 CPU around 50GB of RAM, it's pretty modern so it might be a good example, —Preceding unsigned comment added by 62.30.156.106 (talk) 16:47, 13 June 2008 (UTC)

More Depth Please

I was still left with a few unanswered questions - mainly, simply "Why?" - the answer I assume is performance related, but it would be nice to know more about that - e.g. optimum number of machines in the cluster, How performance increases per machine etc... — Preceding unsigned comment added by 84.189.184.57 (talk • contribs) 22:22, 28 March 2006

optimum number? it depends on the task... if it is really large and easily splittable then as many computers as you can link up. if the task is small or they can't be split, then just one computer will execute best. so basically, if you have a lot of nodes in the cluster then chance is the optimum number of nodes will be reached. note also, you cannot really 'overshoot' as tasks are divided in such a way as to avoid this.

performance doesn't increase normally per machine. the whole system's maximum performance can be much better (as a complicated sum based on processing power but limited by memory and network delays)

As for "Why": yes, any task (computation) that requires a 'supercomputer'... eg: large prime finding, astronomic calculations, clustering (the other type, as in clustering vectors) or ray-tracing and many other graphics computations, or just to run Oblivion with milisecond load time. :D

--Carbon Rodney

Why? Consider this: you can buy a 3GHz PC for say $1000. You cannot buy a 6GHz PC, but you could pay much more than $2000 for a liquid-nitrogen cooled supercomputer. However, you can buy two 3GHz machines for $2000, network them to work together, and for a distibutable application you get the effect of 6GHz cheap. Scientific laboratories do this on a large scale, see e.g. High Performance Computing (HPC). One fun game is to maximize TeraFLOPS per dollar.

And Carbon, how do you load Oblivion on your Beowulf? WINE? Could Vanguard have less lag if I loaded the client on HPC? Pete St.John 19:25, 2 February 2007 (UTC)

If you can get some form of Linux on which WINE will run for the cluster, then you have a machine on which you can game. With enough processor power, you would not need a GPU, because software rendering would be fast enough. —The preceding unsigned comment was added by 150.101.102.188 (talk) 01:49, 3 April 2007 (UTC).

In fairness clusters are NOT at all suited to real time graphics, in fact it's almost a perfect example of how spending a lot of money on a single system is a much better solution. The reason HagermanBot is the how quickly two compute nodes can transfer information - passing data from one computer to another via a network (even if it's a particularly low latency example) is MUCH MUCH MUCH slower than an internal CPU - GPU - RAM configuration in an INDIVIDUAL machine... Clusters are awesome, but don't kid yourself into thinking gaming is an application for them. —Preceding unsigned comment added by 62.30.156.106 (talk) 16:41, 13 June 2008 (UTC)

The Oblivion example was just a joke :) Be aware though, that a Beowulf cluster with 2x 3GHz machines will only approach 6GHz computation power in the absolute best case scenario. That is, the task can be split into two completely independent streams, and there is no need to communicate between the two machines, and the preprocessing and postprocessing time and resources are negligible in comparison to the tasks --Carbonrodney (talk) 08:24, 21 July 2008 (UTC)

Nearly unlimited

What is nearly unlimited? 10^10, 10^100? Non-sensical statement — Preceding unsigned comment added by 86.136.83.170 (talk • contribs) 12:22, 13 July 2006

People usually mean by that, "most folks won't run into significant limits during ordinary use". If I say "a claw hammer lasts forever" I don't mean that it will outlive the Big Crunch, only that carpenters don't worry about the Mean Time To Failure of claw-hammers, but about other things more urgent. It's OK if we use conversational idiom when it won't be mistaken for precise science Pete St.John 17:19, 24 October 2007 (UTC)

Home "Build"

I reverted the caption of an image from "low cost home built" back to "home build". Two things: first, "build" is used by engineers as a noun as in this case, meaning "the built thing" (as opposed to "building" which connotes something large like a house). Also, "low-cost" is redundant with the definition of Beowulf. I don't like to be so picky (there are plenty of editors around who specialize in that :-) but since both elements of the change were a bit weak I just reverted.Pete St.John 16:24, 6 August 2007 (UTC)

On Portal:Free software, Beowulf is now the selected article

Just to let you know. The purpose of selecting an article is both to point readers to the article and to highlight it to potential contributors. It will remain on the portal for a week or so. The previous selected article was LyX. Gronky 02:22, 29 August 2007 (UTC)

Things have moved on, as they nearly always do, and the new selectee is the article on the Debian Free Software Guidelines. --Gronky 21:37, 3 September 2007 (UTC)

XGrid Beowulf or not?

Is Apple's XGrid a Beowulf cluster? It's distributed, UNIX-based and FOSS on the underside. Xgrid's page mentions it ran on OPENSTEP as well, so there must have been FOSS implementaiton of Xgrid as well... —Preceding unsigned comment added by 84.92.248.233 (talk) 23:33, 23 October 2007 (UTC)

Beowulf and Windows ??? Is this possible?

hi..!

I read an article on linking Beowulf clusters with Microsoft Windows using a middleware called Active computing interface (ACI). I would like to know, whether it is possible to work on Beowulf clusters with Windows as a platform? and if it is, can anyone tell me more technical details of it? Do we still call it a Beowulf Cluster?

The link for the article is: http://hpcnc.cpe.ku.ac.th/publication/publication/Publication%20Files/LINKING%20BEOWULF%20CLUSTERS%20AND%20WINDOWS%20ENVIRONMENT%20USING%20A%20MIDDLEWARE%20APPROACH.pdf

--- Priyank. —Preceding unsigned comment added by 129.7.84.3 (talk) 23:58, 24 October 2007 (UTC)

Not possible technically, no; Beowulf is supposed to be Open Source, but MS has gotten into the business of cluster computing. You can read MS's advertising for it at MS's Computer Cluster Server page. Labs which are funded by MS use it. Pete St.John 16:47, 25 October 2007 (UTC)

A Wisconsin undergrad link

I deleted an external link to an undergraduate's blog about his intention to build a beowulf. It looks like he's doing a good thing, but it will be much more useful once they've actually built something and have some experience to write about. Also he might want to find an English major to help him spruce up the verbage, if he wants to meaningfully share his experience. So I wish them luck but nothing for us just yet. Pete St.John (talk) 20:53, 17 January 2008 (UTC)

OSCAR

OSCAR is not an example of MPI software. OSCAR is a set of packages that are useful in setting up and managing a beowulf-type cluster. It includes several MPI packages, but you are not required to install all or any of them. —Preceding unsigned comment added by 12.157.240.2 (talk) 02:21, 19 November 2009 (UTC)

Designed to run Beowulf cluster?

This is mentioned under Operating systems and I found it rather silly because you can setup an MPI2 Beowulf cluster on pretty much any POSIX operating system that can compile the MPICH2 package, or any other such package. So for example I have a Beowulf cluster on FreeBSD but FreeBSD is not listed. I just don't see how Dragonfly BSD would be "designed" to run a Beowulf cluster. — Preceding unsigned comment added by 83.209.23.234 (talk • contribs) 14:29, 25 April 2011

Definition

The definition seems like a long quote. I feel like it could be better understood if someone goes through each of the paragraph and simplifies it in their own words, still keeping a reference. Anyone up for it or have any objections? Cleanelephant (talk) 22:47, 14 December 2011 (UTC)

Article is not that incorrect, but needs touch up. I will do some touch up, add a ref to Sterling's book, etc. History2007 (talk) 11:06, 11 April 2012 (UTC)

Around 1998 I was building a commercial laser-printer raster image processor (RIP) consisting of five (5) COTS motherboards connected by Ethernet, and built on Beowolf ideas. At the time, a key technology was to bond several Ethernet channels per processing node into a higher bandwidth connection. As I was not doing this, there was an internal question of the legitimacy of calling our system "based on Beowolf". Is an enhanced communication channel part of the Beowolf definition? foxkid (talk) 15:27, 25 February 2016 (UTC)

Hypercube

Back in the mid-80s I did some computer work at JPL. At that time I heard some of the other computer guys talking about setting up a hypercube. When I asked what that referred to, I was told that it was a form of cluster where the computers were (conceptually, not literally) set on the vertices of a hypercube and only capable of communicating with each other along the edges. Assuming that the proper software is used, is it possible to make one that's also a Beowulf cluster? If so, would it be most appropriate to mention it here, at Hypercube internetwork topology or both? JDZeff (talk) 00:26, 1 February 2024 (UTC)