2.2.1 SMP Scalability to 4-way

urbanski@us.ibm.com
Sat, 30 Jan 1999 23:29:26 -0700

OK this is a long story but I'll try to be brief. We were asked to
benchmark a computational chemistry application for a major drug company on
a 4-way Xeon machine. They wanted Unix, and the only two supported on X86
were Solaris and Linux. The SMP performance of the 2.0.x kernels was poor,
so we were forced to bid Solaris on this one - shame because it would have
been a good reference - they need 36 4-way Xeon machines to meet their
computational requirements.

But in any case I wanted to try Linux again when 2.2 came out - so I just
re-ran the numbers. It's decidedly a good-news, bad news case. Here they
are (times are in seconds)

Linux 512M RAM Scalability (from 1X)

1 CPU = 1503 -
2 CPU = 702 107%
3 CPU = 494 101.4%
4 CPU = 954 39%

Solaris 512M RAM

1 CPU = 1029 -
2 CPU = 588 87.5%
3 CPU = 447 76.7%
4 CPU = 411 62.6%

Linux scales much better up to 3 CPU's - so much so that we almost caught
up at the end - in fact we might have because the Solaris machine is using
1M cache CPU's compared to the 512K on the Linux box. I'll do an
apples-to-apples comparison tomorrow. But there is obviously somthing
wrong when I go to 4 CPUs - it takes longer than 3, or even 2! This is
discouraging because I have several other opportunities but I can't
recommend Linux for them until I know it can scale reliably. If anyone can
help me troubleshoot this please copy me via email.

Jay Urbanski
Netfinity Systems Engineer
IBM Advanced Technical Support
MCSE, PSE, Certified Solaris Systems Administrator
(817)962-3597 TL 522-3597
(817)962-7307 fax
(800)413-9093 pager
urbanski@us.ibm.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/