[JamesY]

Anyone else here using openMosix or tried it out?

I haven't looked into it to far but I was planning on using clusterKnoppix to build a little minifarm in my room so I thought I might aswell get asking.

Does it work on FreeBSD? Would MUCH rather prefer a minimal install of FreeBSD then load up openMosix and do it that way, alot less CPU/MEM usage, good for older computers as clusters.

Simplicity of running programs on the farm?

Main question, if I do a minimal install of say Freebsd, and load up openMosix, get everyone connected up and registering and such, inorder to use say, Blender to render 3d images, do I have to have Blender installed on EVERY box, or does it basically just send info to the CPU/Mem/etc... and send it back to the main machine? Basically FreeBSD+openMosix = compatible with running anything main can, OR, FreeBSD+openMosix+other software such as blender on ALL machines.

Thanks

[horndude]

Yes,ive tried it.

no you dont need the app your running installed on all the machines,just one, openmosix will migrate the processes to the other machines on its own

biggest thing is each machine must have identical kernel series with the open mosix patch compiled in, add the openmosix filesystem sharing option too, it will help

also, latency is a serious issue, network will work way better with a switch vs using a hub

I would assume it will work on BSD, its a kernel change, nothing more than that really, there's some other monitoring software with it and provision to turn openmosix on and off and thats about it, dont see why it wouldnt work on any of the BSD's too work on

you can set it up to automatically have each node connect and add itself to the cluster when booted up

my experience with it showed it works, it works better with each node being close in performance,and it works better the more nodes you have

[JamesY]

Well I got the switch in my room, and enough 10/100mb cards for the computers right now and it wont be hard to get more so 100mbit switched network in my room should work fine.

I thought the reason openMosix was different then from say a Beowolf cluster was that the machines and kernels didn't have to be identical so it was more universal then a Beowolf cluster would be.

The nodes would be close in "performance", being around 200mhz isnt exactly performance , but it is basically just a learning tool right now, if I enjoy it and use it alot for folding or other things then I will be slowy upgrading the cluster to newer machines that I get given.

[C.Ingram]

openMosix is Linux-only. It is a kernel extension that migrates processes between nodes. Therefore, you'll need all Kernels to be identical to make sure you don't have any issues.

Since you are doing this in your home, don't worry to much about anything. If you were going for an academic or pro (or personal heavy use cluster) you would want a dedicated 100/1000 network for the nodes to talk to each other with nothing else in the way.

Make sure whatever you are clustering forks multiple processes. Since openMosix works be migrating processes you will need multiple worker processes to see any benifit.

[JamesY]

Ok so I would have to be running say all one clusterKnoppix distro to make sure I wouldn't have an issue, that makes sense I guess.

Dedicated network, yah for a first cluster/messing around, 100mbit should do me fine, I could probably pickup a switch but the cards would be the thing that would do me over

So by the process thing you mean I would need something that ran kind of like Apache's httpd, multiple processes of the same thing running at the same time? So Folding@Home wouldn't work?

[C.Ingram]

Originally posted by JamesY
So by the process thing you mean I would need something that ran kind of like Apache's httpd, multiple processes of the same thing running at the same time? So Folding@Home wouldn't work?

That is one of the ways Apache can work, yes. That wouldn't be a good idea, though, because disk I/O requests would just wind up getting routed back. Try and take the data you need with you in process memory or hit a 3rd party source of data.

Folding at Home might work. I belive each instance spawns a few processes, but they may need each other. In which case you wind up sending IPC back and forth killing some of the speed benifit. It should still help though.

[JamesY]

Guess I'll just have to give it a shot and we will see.

Mainly I was trying to figure out....
F@H Spread out over the cluster = faster at crunching
or if..
F@H running independantly on each machine would be faster, I mean 200mhz crunching numbers still takes about 4-5 days PER unit , pretty sad so I thought maybe I could speed it up.

I guess the whole project is kind of a learning process

[C.Ingram]

I really couldn't tell you. Folding@home creates a hetergenious cluster across the internet. In the case of openMosix, I dare to say you should see some improvement. Would it be better than running seperate clients on each machine? I don't think so.