OpenMosix is best on isolated, relatively homogenous clusters. It will work otherwise, but is fraught with problems. It will not migrate gcc compiles for instance as most are short lived processes. Then it can migrate things like ssh console sessions, so that when you shut a system down, it will kill your session when migrating the process back. If you use very unbalanced hardware (e.g. new athlon vs old cyrix) it can take a very long time while waiting for the slowest processor. You can configure around this, but its time consuming.
Security, better these days but still not something you want running on a gateway. note that the stable OM kernels are before the brk() patches. Often (very anoying this) one node will take down every other node on a cluster - either a full hang or spontaneous reset. There is an openmosix mailing list which is quite good. Most of the problems with 2.4.23 have been fixed and I am waiting for gentoos sources to pick up on them. Suitable applications for OM are those with no shared memory (there is a patch, but it seems experimental and problematic), relatively small data requirements (there are shared file systems, but that appears to have one of the current crop of bugs) and having long lived, processor intensive jobs. Its worth trying just from the experience point of view: you soon develop a healthy respect for the abilities of someone managing hundreds of these beasts! - and get a first hand understanding of the difference between hype and reality - the reality is the one that bites! BillK On Wed, 2003-12-24 at 11:29, Brett I. Holcomb wrote: > Okay. I haven't been there but have been thinking about it <G>. I tried to > check out openmosix but the kernel version they support doesn't have XFS > built in so I have to wait until they decide to support 2.4.23 which does > have XFS - in short: I gave up on openmosix for a while. > > You say it doesn't work and mention the bugs - can you tell me more before I > try it. > > Thanks. > > On Tuesday 23 December 2003 21:59, you wrote: > > been there, done that ... > > > > OpenMosix is actually severely limited in the real world. Even with the > > shared mem patch (I am not sure if the gentoo emerge adds this) it wont > > work with mysql. And there's the fact that the current 2.4.23 is very > > unstable oM probs), and the pipe bug, and ... > > > > > > Large databases must be able to loadshare somehow. > > > > BillK > > > > On Wed, 2003-12-24 at 11:01, Brett I. Holcomb wrote: > > > You might check out openmosix - it's in portage. > > > > > > On Tuesday 23 December 2003 21:27, you wrote: > > > > I run a mysql database on a relatively low powered gateway (cyrix233) > > > > and find that the athlon I ran it on before was far faster (of > > > > course!). Is there a way to loadshare using mysql so that when the > > > > athlon is up, it takes the load, and the gateway otherwise? > > > > > > > > I guess some kind of replication and and load distribution overlay is > > > > required. I am sure that this has been solved elsewhere, but how? > > > > > > > > BillK > > > > > > -- > > > [EMAIL PROTECTED] mailing list > > -- > [EMAIL PROTECTED] mailing list -- [EMAIL PROTECTED] mailing list
