[K12OSN] $5,000 cluster server powers 120 clients?

[Robert Arkiletian]

Petre Scheie wrote:

> Depends on what you mean by 'cluster'.  If you're talking about using 
> OpenMosix to cluster them, then no, I don't think it's a good 
> approach.  We had some discussion of this a while back, and the 
> problem is getting applications to migrate from one server to another 
> as needed; in short, it doesn't work with things like Mozilla & 
> OpenOffice.
>
> The approach that I think makes the most sense is to use one or two 
> servers for just providing the KDE/Gnome/IceWM environment, and then 
> dedicating the other servers to one or two applications, ultimately 
> with one app per server.  That way, pigs like OOo don't affect the 
> performance of lighter faster apps.  Also, to upgrade an app, you just 
> get it going on a new server, once it's ready, you change the link on 
> the users' menus/desktops, and there's no downtime.
>
I thought a cluster would parallelise processess. After reading the faq 
for OpenMosix I see the problem. It's a High Availability cluster as 
opposed to a High Performance cluster.

--------------------------------------------------- snip 
---------------------------------------------------------

*What is openMosix not useful for?*:

openMosix lets a cluster of computers behave like one big 
multi-processor computer. However, it doesn't automatically parallelise 
programs. Each individual process only runs on one computer at a time. 
For example, if your computer could convert a WAV to a MP3 in a minute, 
then buying another nine computers and joining them in a ten-node 
openMosix cluster would *NOT* let you convert a WAV in six seconds. 
However, what it *would* allow you to do is convert 10 WAVs 
simultanously. Each one would take a minute, but since you can do lots 
in parallel you'd get through your CD collection much faster :).

If what you need to do is take a single process and parallelise it 
across multiple machines, then openMosix is probably not the technology 
you're looking for.

Mosix works best when running plenty of separate CPU intensive tasks. 
Shared memory is its big drawback, like *Beowulf*: for applications 
using shared memory, such as Web servers or database servers, there will 
not be any benefit from [Open]Mosix because all processes accessing said 
shared memory must resided on the same node.

--------------------------------------------------- snip 
---------------------------------------------------------


Oh well, it sounded like a good idea but I don't think all the apps in 
fedora are going to re-written for a beowulf cluster anytime soon.


> The proof-in-the-pudding for this approach is Largo, FL, where they 
> use this exact approach.  Doing so allows them to support 240 users 
> with a couple of 900mhz terminal servers.  They found they could get 
> the per-user memory consumption down to about 13M and I think that was 
> using KDE. (They don't use LTSP, preferring to roll their own terminal 
> servers.)  Also, with this approach, you spend money only on the apps 
> that need a lot of resources, e.g. more memory for those apps that 
> need it.  It also scales up better because for each additional app 
> that you add, you just add another server.  But, if money is tight, 
> and, say, you can only afford four app servers but you have 16 apps, 
> you can put multiple apps on each server, for an average of 4 
> apps/server, and then next year get just one more server, move some 
> apps to that new server (probably a newer version of the app in 
> question), for an average of 3.2 apps/server and repeat in the years 
> after that.  In some cases those servers may not even need to be that 
> big, depending on the apps.
>
> Petre
>
Sounds interesting but I think I'll stick my dual Xeon and k12ltsp. 
Rolling your own Terminal Server does not sound easy. Now I understand 
why I did not get many bites on this thread. Thanks for the info Petre.

Robert Arkiletian