Torture-testing a Linux Cluster...

[Entity]

I don't know how many of you are familiar with Mosix at all - it's an app that we're using here at work, for creating a linux cluster. Anyway, we're trying to torture test it...right now we've got 11 nodes running off of it, with around 7.5ghz of power, and we want to test this setup. We tried loading it down with RC5, but for some reason, it doesn't properly distribute the processes - if we instruct it to manually, we can get it to assign the processes to different nodes; but for some reason, it doesn't do it automatically (like Mosix does with other processes).

Basically, we're looking for something to run to torture-test/benchmark the setup, to see how it's running right now. Anything we've thrown at it, the machine handles without problems...so what can we do to mess with it?

thanks,
Rob

 

[networkman]

How about multiple instances of SETI?

 

[Entity]

Seti would work, I guess, as long as we can get it to properly distribute across the system.

Is anyone here familiar with Mosix?

Rob

 

[TwoFace]

I think Kilowatt is familiar with Mosix, I'm not I just know what it is

Just because I know I'll be interested in the answer (dreaming about the day I can build my own really) I thought I'd bump this up even tho' it isn't even on page 2 yet...

Be sure to tell us how it handles itself Entity, and good luck!

With love and respect your fellow TA member

Two-Face

 

[Poof]

My co-worker runs Mosix on his home LAN - about 5-6 nodes out of about 11 available, are on it right now...

He also ran seti on it at one time, although seti is not multi-threaded (so a single process won't run across the cluster).

What he did discover though, was that since his nodes consisted of all different types of processors, if he ran a seti instance on each node and set Mosix up for dynamic (rather than fixing processes to a node), and if he happened to telnet to a node to do something, the seti process on that machine would move off of and temporarily run on another node, and would then come back to its originator node when he was done messing with that machine...

Interesting stuff...

 

[TwoFace]

^TTT /me still wants to know

TF

 

[Entity]

^ Still looking for a good answer...

Rob

 

[RaySun2Be]

Entity,

There was a recent discussion of Mosix where they got RC5 to work, they had to change some config parms. Mosix & RC5

 

[JHutch]

Mosix has always sounded like one of those really cool things that I've never had time to learn ...

Thanks for the info (and the link to the other discussion, Ray). Even if I don't have time to actually play with it, this stuff is always interesting to read about.

JHutch

 

[Poof]

JHutch - it seems to me that with Mosix, there really isn't much to "learn". You substitute your current kernel with the Mosix one, setup the map files on each node, and away you go. For me, I would need to upgrade a couple of machines to the newer libraries to make use of it on more than maybe 2 machines that stay booted in Linux (and there's also a limitation to only working on x86 platforms, which is another problem I have), but it appears to be almost idiot-proof as opposed to some of the other clustering software solutions out there...