• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Server Farm & Clusters

jose

Platinum Member
I'd like to learn how to setup a Server Farm "Computational Clusters". Are there any books on this subject ? (More on implementation, not theory) I'm familiar w/ Linux & programming (BS Computer Science), but to date spent my time on developing vertical apps. (Enterpise Databases w/ gui front ends) I'd like to learn more about server farms. ie how are jobs submitted/processed, etc.

Do you have to write your own job control language to interact w/ the various languages available today ?

Thank you for help/info.

Regards,
Jose
 
Originally posted by: jose
ttt

Anybody out there ??

Regards,
Jose


Well from what I understand you have 2 types of Linux clusters that are commonly delt with.

You have a OpenMosix cluster. This is the easier one to setup and run, there are even some Knoppix-type distros created to make it easy to setup nodes.

Basicly how this one works is that you create a master and some nodes. What it does then is migrate proccesses/threads back and forth between all the computers. Say your going to encode rip a couple DVDS, so you setup your cluster copy the DVD to disk and write a script to encode those DVD's using Mplayer's Mencoder and libavcodec's mpeg4 codec (Divx clone, but better). So say you want to rip 4 at the same time using very high quality and high resolution. So you start them off all on the same machine, they all startup and bog the machine down. After a few seconds OpenMosix will kick in and migrate those proccesses/forks/threads/whatever to different machines based on what resources they have avaible. Faster machines get higher priority, machines with less CPU/Memory usage gets higher priority, etc etc.

Pretty simple stuff, the only gotcha is that this is happening at the kernel level so you have to be aware of what CPU is capable of. So if your using Gentoo and have mencoder compiled and optimized using SSE (for instance) instructions and you migrate it to a machine that doesn't support SSE, then it's likely your going to have a crash. Also they all have to be running the same OpenMosix version patches, not neccirarialy the exact same kernel, maybe kernel version though.

The other is Beowolf. Beowolf is about number crunching machine, big managed batch jobs and stuff like that. OpenMosix is nice because it works with all applications, it's mostly like running a SMP machine. Beowolf is about just plain old computational cluster.

Commonly used for things like analizing the structure of DNA, weather patterns, protein creation, drug creation, astronomy. That sort of things, any time you just want to crunch numbers.

They have libraries and stuff setup for you to help build applications, and there are a few different types of structure you can use. Not to familar with programming stuff and I don't know the details about this sort of thing.

Also you can mix and match. You can use a OpenMosix cluster on a Beowolf cluster. May help with optimizing things, like say you have a Beowolf cluster that has different speed machines and you have a job that requires all the jobs to get finished before going onto the next step. So the fast machines get done and the slow machines are chugging away, OpenMosix will migrate the threads from the slow machine to the faster ones once they are idle. So that way you can aviod having to deal with a complicated scedualler, or something like that.

Very fuzzy on the details, so don't listen to me to closely.

Beowolf dot org
Links to mailing lists and stuff like that.

Home of openmosix.
 
Back
Top