[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On Wednesday 25 June 2008 07:06, tom wrote: > Grant Sewell wrote: > > Hi all, > > > > Whilst watching a foundation degree student's presentation on their > > project - building a Linux beowulf cluster - the idea struck me... what > > a wonderful way to generate enough processing power for an LTSP server. > > > > I know there are some people on the list who have a bit of LTSP > > experience, so I was just wondering if (in their experience) a beowulf > > cluster would actually work well as an LTSP server? If it were to > > work, it could even benefit from being configured as a COWs (Cluster of > > Workstation) - so the LTSP server cluster would also be able to pull > > processing power from the idle workstations connecting to it! > > > > It's late and I'm getting excited about this now... dammit, I'll never > > sleep! > > > > Cheers all. > > Grant. > > Beowulf clusters are about clustering machines that can "kick butt" > processing wise. LTSP is more about centralizing resources (reducing > admin time) and allowing lower spec machines to be kept in service > longer as clients. > > Someone designing a computationally intensive cluster of computers would > normally want to use the fastest processors available, and standardise > on machine build. > > While in theory your idea is sound. I think you would want to consider > the cost in terms of power consumption / processing power. You could end > up with a very powerful (but very expensive to run) cluster. > > Also consider that there is a scaling factor, you would need to add more > lower spec machines to the cluster to get the same power, and each > machine needs to be synchronised with the rest of the array, the law of > diminishing returns kicks in quicker the lower the spec of the machines > are. > > You can't get more out than you put in, so if everybody on the cluster > wanted to run kick butt processes, the performance would degrade quickly. > > On the other hand it would be fun to try it, as long as I don't have to > pay the energy bill ;-) I thought the idea was, essentially, to use the computational power of the 'Terminals' (in LTSP) as part of the server. Good point about the lower spec machines - all Beowulfs I've seen are collections of nearly identical machines. However you can easily envisage asymetric machines running bits of a partitioned project: Many moons ago I had a sort of cluster of machines processing a POV animation . A 'controling' machine managed the job and other machines came in and took small timeslices of the animation to do and returned the processed images to the controller and got new work until all was done. There was a 386 in there which was a router but it still managed to process a few images while the pentiums were doing hundreds in the same time. This concept can easily be extended to most types of jobs. Tom te tom te tom -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/linux_adm/list-faq.html