Comments on: Beowulf or Mosix – solving the parking lot problem through simulation

By: Docs.Rage.Net

Docs.Rage.Net — Mon, 30 May 2005 07:28:43 +0000

Docs.Rage.NET: /howto/korean/Beowulf-HOWTO.html

This was suggested as being relevant by a visitor.

By: Kevin Hobson

Kevin Hobson — Tue, 22 Mar 2005 07:11:49 +0000

I know I may be treading into Proverbs 30 territory (give, give), but…

If someone does happen to develop an open source church resource management application, I would sure love to see volunteer management added to the mix. And here’s the kicker: when your volunteers are working in the nursery (where there is a requirement for adult to children ratios based on the age group) and some are parents who want to be in the same room as their child or are a husband and wife and they want to specifically share time, or specifically not share time (one or the other is with their own children) or one is in the choir which blocks out the beginning of both services, etc. Anyway, there’s a combinatorial nightmare for you I have yet to find an application to address.

By: Ted Carnahan

Ted Carnahan — Mon, 21 Mar 2005 04:16:53 +0000

I strongly recommend the Bootable Cluster CD. It can be burned on a business card CD, carries MPICH and LAM (the two leading open-source MPI (Message Passing Interface) distributions), boots very quickly, is lightweight, etc. It doesn’t have as much “stuff” as ClusterKnoppix or the like, and it’s not a single system image like Mosix, but as far as using the resources most efficiently, it’s quite nice.

Having done a little high-performance computing (primarily for chemistry), the biggest bottleneck for networks of this sort is communications latency. That’s why the big Origin and Altix supercomputers (and even home-grown solutions using Myrinet) specialize in being low-latency. It won’t matter for “loosely-coupled” problems, but for anything “tightly-coupled” (i.e. requiring a lot of data exchange) latency is the bottleneck, and unless you’re using a LOT of cpus, it’s a waste to even start programming in parallel.

Nowadays, Sun Microsystems and IBM are rolling out pay-as-you-go supercomputing, with fairly low cost per Service Unit (typically a single CPU for an hour, costing ~ $0.80 to $1.00). This is more likely for serious problem solving, but is unlikely to be available to smaller buyers for several years.

Something else to remember – computers use electricity and require cooling.

Okay, practicality aside, I LOVE the idea of having my own cluster. It’s a tremendous feeling to control that much computing power. I’ll never forget when I first logged in interactively on an Altix system, typed “top” and got 64 lines of CPU summary. It was sweet. I geek out on that sort of thing.

For small projects of this sort, the bootable cluster CD type projects are the way to go. I really love that flexibility. Anyway, enough rambling.