[Beowulf] Rackable / SGI

[Chris Samuel]

----- "Mark Hahn" <hahn at mcmaster.ca> wrote:

> unclear in several ways.  for instance, linux has hotplug cpu
> and memory support, but I really think this is dubious, since 
> there's damn little hardware that supports it, _anywhere_.

It works on our 4 year old IBM Power5 cluster, if the ASM
detects CPU issues it can get the kernel to hotplug that
core out of the system.    Doesn't need to be a physical
removal to be useful!
 
> XFS may have been fairly "enterprise" for its
> time - it's been  available for linux for quite
> a while, I think.

First code drop from SGI in 2000, Mandrake 8.1 had
native support in 2001.  Christoph Hellwig has a bit
of a history of it in his slides from UKUUG in 2003.

http://oss.sgi.com/projects/xfs/papers/ukuug2003.pdf

> but if you look at options today, is it clearly the only
> "enterprise" choice? certainly not - ext3 and 4 are
> certainly viable,

I wouldn't let ext3 be on an NFS server that's getting
much load, at LCA2005 in Canberra one of the papers was
on scalability in Linux and they commented that ext3 in
2.6 was single threaded through the journal daemon, so if
you had a lot of writes they'd all back up on one another
in the queue.

I suspect that the default journaling mode for ext3 will
change from data=ordered to data=writeback in 2.6.30 will
help there, as will some of the other bug fixing documented
in the LWN article "Solving the ext3 latency problem" which
is here http://lwn.net/Articles/328363/ (subscribers only
for the next few days - please support LWN if you like what
they do).

I think ext4 is still a bit too new to inflict on others,
though I'm happy enough to use it for /home @home.

> going forward, I expect to use ext4 and probably btrfs;

I think btrfs is very interesting, but again I reckon it'll
be quite some time before it'll be ready for production use.
I've been using it on my laptop for a while now and it's not
broken anything (yet), which is fairly impressive.


> cheap shot - don't you remember that linux was 64b quite early because
> DEC dropped an alpha on Linus?

DEC also had their own separate porting effort to the Alpha too,
and Jim Paradis (who started that) writes about that in this 1995
Linux Journal article:

http://www.linuxjournal.com/article/1044

# Our involvement in Linux began at the end of 1993, when we
# realized that there was no entry-level operating system for
# Alpha-based systems.

>  it was also SMP-aware pretty early.

For Intel and Sun yes, but not Alpha..

The work was helped because SCO, er, Caldera, dropped
an ASUS SMP motherboard on Alan Cox (which they conveniently
forgot about before suing IBM)..

http://web.archive.org/web/19971211104421/http://www.linux.org.uk/SMP/title.html

# At the moment I am developing using a Caldera
# provided ASUS P54PNIP4 motherboard with 16Mb
# of RAM.

cheers,
Chris
-- 
Christopher Samuel - (03) 9925 4751 - Systems Manager
 The Victorian Partnership for Advanced Computing
 P.O. Box 201, Carlton South, VIC 3053, Australia
VPAC is a not-for-profit Registered Research Agency