[SGVLUG] questions about linux clusters

[Ted Arden]

On Sun, 4 Dec 2005, jmd wrote:

> i am building a linux cluster for fun/home use. not for any sort of
> production. has anyone worked with them?

sure...  built a 'few' in my time.

> the book i am reading about clusters seems to suggest that the only
> problem i may encounter is when hardware varies and disk images must
> be created for various hardware in each node. i plan on using rocks
> for the OS which requires a disk image on each node.

NPACI ROCKS is pretty cool. worked with some of the
guys down there at the san diego super computer center
that started it.  it's basically a kickstart based
ditro method (although that could have changed, been
a while since i used it).  you may also want to look
into systemimager (been using that since it was
developed at VA Linux.. worked on some of it)

i guess the bigger question is, what are you going to
be doing with this cluster besides burning electrons?
job scheduling?  paralized compiling? lottery number
generation??  hahahah.

> so my question is, will the disparity of the systems speed become a
> problem if i use the faster dual xeons as a head node.

depends on the app. using pbs or something, it'll
just farm jobs to the freed resource.

> i realize
> this will depend in part on the given application. since this is a
> learning experience, obtaining another eseries that matches the other
> 2 and skipping the faster board is ok with me.

you may wanna think about getting a gig-nic.
will improve things quite a bit.  nics/switches
are pretty cheap right now.

> at present my hardware is simple: 2 dual cpu ibm x330 servers and
> possibly the asus dual cpu xeon. hard disks are 160 scsi, and
> ethernet is 10/100.

=ted=