NUMA, Sequent NUMA-Q, etc

NUMA, Sequent NUMA-Q, etc

[Kevin Bowling]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1418 bytes --]

NUMA is something that's been on my mind a lot lately.  Partially in
seeding beastie ideas into Larry McVoy's brain.

I asked Paul McKenney for some history on what went down at Sequent
since that's before my time.  He sent me this, which I think the group
will enjoy:  http://www2.rdrop.com/users/paulmck/techreports/stingcacm3.1999.08.04a.pdf

It looks pretty nice.  Not sure anyone's come as close as Irix to
solving and productizing "easy" NUMA but that's the one I have the
most hands on experience with.  They can affine, place, migrate, and
even replicate many types of resources including vnodes.  I'm actually
surprised all that code seems to have been spiked and it doesn't seem
like either Sequent née IBM nor SGI brought forward any of their
architecture to Linux.  Paul did RCU which is a tour de force, but the
 Linux topology and MM code looks like the product of sustaining
engineers instead of architectural decree.  Maybe the SCO lawsuit
snubbed all of that?

HP has an out of date competitive analysis that's worth a look
http://h20566.www2.hpe.com/hpsc/doc/public/display?sp4ts.oid=5060289&docLocale=en_US&docId=emr_na-c02670417.
I don't have enough seat time with Tru64 but maybe they had some good
ideas.

As open source, I do like Illumos' locality groups.  I can't make much
sense of Linux on this, too much seems to be in arch/ vs a first class
concept like locality groups.

Regards,
Kevin