High performance cluster

Hello and thanks for reading this!

I want to set up a cluster system containing of 6 dual systems
(2x 1Ghz, 1Gb RAM each). The whole system (final result) is going
to run a high traffic website with intensive database connectivity.
(Apache+MySQL).

Untill now i was able to find 2 solutions:

-Heartbeat (www.linux-ha.org)
-GridEngine of SUN

Four questions:

1) Which one of these solutions are going to output most
performance for the mentioned configuration?

2) Which is easier to setup?

3) Do i have to setup a virtual networking mechanism or the
local 192.168.0. ... numbers will do it?

3) Shouldn't distributed computing be handled at Kernel level?
As far as i understand both solutions are userland software and
if it is the case that they work with message exchanging, how does
that stand compare to true low-level communicaion?
(This is a very general question. just something i wondered, you can
igonre it.)

Thanks a lot for your time and suggestions,
kave

Hi,

(E-Mail Removed) wrote:
> -Heartbeat (www.linux-ha.org)

this will give you no performance at all. It's a failover cluster only.

cheers, michel

(E-Mail Removed) wrote:

Note: multi-posting is frowned upon ;-(

> Hello and thanks for reading this!
>
> I want to set up a cluster system containing of 6 dual systems
> (2x 1Ghz, 1Gb RAM each). The whole system (final result) is going
> to run a high traffic website with intensive database connectivity.
> (Apache+MySQL).

How did you come to these resourse numbers?

> Untill now i was able to find 2 solutions:
>
> -Heartbeat (www.linux-ha.org)
> -GridEngine of SUN

Grids and clusters are not the same. I suspect you want to look at
clustering. [Note that Heartbeat is just one element of a cluster.]

> Four questions:
>
> 1) Which one of these solutions are going to output most
> performance for the mentioned configuration?

Depends on what you mean by "performance". And "what" configuration?
What setting will it run in? Internet or intranet? Expected size of
db? What's your main concern -- traffic load or complex data retrieval
or what?

> 2) Which is easier to setup?

Only you can answer that. What are you already familiar with?

> 3) Do i have to setup a virtual networking mechanism or the
> local 192.168.0. ... numbers will do it?

Huh??? "Virtual networking"? 192.168.x.x is a private IP address.
Might work on your private lan. Internet access will require a public
IP somewhere :-)

> 3) Shouldn't distributed computing be handled at Kernel level?

Huh??? The whole point is to abstract away the hardware _and_ OS
specifics as a general rule.

> As far as i understand both solutions are userland software and
> if it is the case that they work with message exchanging, how does
> that stand compare to true low-level communicaion?

It abstacts away the low-level details and allows _different_ platforms
to participate. _Compute_ nodes in a grid designed for a _specific_
purpose (usually involving parallel processing) often use
tuned/stripped down kernels.

> (This is a very general question. just something i wondered, you can
> igonre it.)

Curiosity is good. Google is your friend

http://www.redhat.com/docs/manuals/csgfs/
http://www-128.ibm.com/developerworks/grid/
http://www.globus.org/solutions/
http://gridengine.sunsource.net/

Again, clusters and grids are different animals -- though clusters are
often part of a grid.

> Thanks a lot for your time and suggestions,
> kave

Your welcome.

prg