Clusters (was CLUG meeting 23 May 2002)

[Simon Haddon]

>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<

On 22/05/02, 21:43:00, "Richard Cottrill" <richard_c at tpg.com.au> wrote 
regarding Clusters (was CLUG meeting 23 May 2002):


> > Same as Matt I have not been able to attend for a while now.  If anyone
> > is interested (and I can make it) then I could talk about my experience
> > in setting up a Linux/mosix cluster with failover for web hosting.

> I won't be able to make to any meetings for the foreseeable future; but I 
am
> curious about how well this sort of thing works. The OpenMosix web site
> suggests it's not really suitable for shared memory applications
> (specifically citing web servers).
You right that at this stage it does not handle shared memory threads but 
that will change in the future.  The main benefits that it has given me 
is that if the main server is too busy and I want to SSH to the box then 
mosix will kick in and move my login to another box (transparently) and I 
can run some of the processes on the other mosix clustered boxes to help 
share the load.  It also provides quite a secure method for sharing the 
disks over the network using MFS (mosix file system) which makes it 
easier for rsyncing between machines and other things.

> Can anyone give me a concise description of what 'grid computing' means?
> >From what I can tell it's essentially a different name for a beowolf
> cluster... I think there's supposed to be something about the nodes not
> being dedicated cluster machines in there too.

> My take on grids is to wonder about the possibilities for running mosix 
as a
> user mode Linux process. I guess there'd need to be some way for the 
mosix
> scheduler/load balancer to account for the fluctuations in the load of 
the
> 'real' Linux host.

> Actually, how feasible is it to port UML to other OSes? Evil thoughts
> occur... :)

> Richard