Most of you might already know the cluster project OpenMosix. OpenMosix has been around for awhile now and has worked with many vendors to create a “patch” for the 2.4 Kernel tree. It also makes userland utilities for this cluster patch. But what a lot of people don’t know is that they have started to work on a patch for the 2.6 Kernel tree. For a lot of users that now use the 2.6 Kernel this gives them the opportunity to run cluster at home or at work without having to buy very expensive equipment. There is one site that now makes it easy to patch your kernel now for the 2.6 Kernel tree. What would be nice is to see this in the main line tree, if anything in the 2.7 tree.
I’ve tried this a few times with the result being a loss in system stability for machines in the cluster.
Hopefully this will go mainline and it can be better maintained since I’m still interested in trying to deploy it. The concept behind it is very good.
Are you refering to stability in 2.4?
I played a little with OpenMosix, but due to the nature of my work (heavy scientifc / numeric computing) and the fact that I have to write most programs myself or use scientific libs, I found MPI (especially MPICH2) way more suitable *for me* than OpenMosix.
However, every progress in the area of Parallel Computing is good, so perhaps I will give it a shot !
You should try running your mpich apps on top of an Openmosix cluster. If your junk is rather cpu bound it should have similar execution times. And depending on how busy your machines are and what your current setup is, you and your fellow users maybe won’t need to deal with things like grid engine anymore, or fight over machines. Plus, if you don’t have tight integration with something like grid engine, it can make cleaning up after dead jobs quite a bit easier. Or anyway, Openmosix is fun to play with at the very least.
I’ve never had any stability problems with Openmosix, and I’ve played with it quite a bit over the last few years. Are you sure it’s not flakey hardware that’s the problem?
Someone please make some userland tools for 2.6
There is userland tools, just look on the openmosix site.
Also in gentoo, theres a few apps for openmosix
now that mosix is GPL’d again, isnt it time to merge the 2 projects?
It might have been the hardware.
2.4.x series, We have a pair of dual athlon MPX machines with dual port gigabit, 3ware and 4 drives. Heavy processor access, heavy IO and network access was normal for these machines. Normal uptimes were about 3 months, with the openmosix it went down to a few days.
Tried before that even to cluster a couple of dual 1GHz P3 boxes, their uptimes went down to about a week, but with plenty of kernel errors.
One big problem is that there really aren’t so many resources available for testing. Also we use these boxes for processing, nfs and samba access in a production environment. I don’t really know of anyone who hits their systems across the board as much as we do.
The userland tools don’t work for 2.6