I'm sick of tossing out old computers, and really haven't ever needed
to use them for low-end projects. Every few years, though, it seems as
if I run up against the limits of the reasonably powerful system I
built a few years back. So I'm thinking of setting up a cluster using
openmosix. I'm very new to clustering, and only recently (yesterday)
started researching it, so I'd like to make sure my reasoning is sound
before I start doing something like this.

I'm currently using my box heavily as a digital audio workstation. I
want to do realtime mixing of inputs, as well as effect processing and
rendering of the finished tracks. Simultaneous compression to Vorbis
and flac would be kinda cool, too. :) Most of my DAW apps are
multithreaded, so I'm hoping to take advantage of process migration
for the realtime work.

So I've been delving into how clusters work. While it'd be convenient
and cheaper to just string together a couple old boxes, I move a lot
and really don't want to lug around 3-4 towers, reconnect the LAN,
etc. So I'm pondering switching to rackmounts. Problem is, I know very
little about rackmounts. I priced some cases on pricewatch, and they
seemed somewhat expensive ($90 for a 1u.) Would rackmounts be the only
way to go for portability? How difficult is assembling rackmounts, or
would it be better to buy them preassembled? Are there any good
venders of inexpensive rackmounts, especially as I'm not necessarily
looking into one massive powerhouse, but several smaller units?

I'm also pondering the logistics of how to allocate resources. My
initial thought was to start with one, extremely powerful node which
I'd install openmosix on at a later date when I wanted to
expand. Then, in a couple years when I'm buying that used 6 GHz box
which I've doubtlessly been eying by then, I wouldn't need to toss out
my old hardware, and when the rack is eventually filled, I can toss
the least powerful node and not feel all that bad. But I'm wondering
if maybe it's better to buy several lower-power systems? Also, reading
the openmosix docs indicates that separate filesystems need to be
installed on each node, though MFS can eventually be used to link
common directories. I can see the advantages of this, but how
difficult would it be to install a fairly massive, fast scsi drive on
the master, while booting all slaves disklessly? Would the MFS/disk rw
hits substantially slow the cluster? It just seems as if administering
every node separately is vastly inefficient in systems where all nodes
are under one person's control.
_______________________________________________
Siglinux mailing list
[EMAIL PROTECTED]
http://www.utacm.org/mailman/listinfo/siglinux

Reply via email to