Numbers are off [LWN.net]

Numbers are off

Posted Aug 21, 2003 15:19 UTC (Thu) by pflugstad (subscriber, #224)
Parent article: User-data replication on NUMA systems

I think you're looking at the numbers wrong. The posting has:

SDET Average Throughput (NUMA-Q):
2.6.0-test3 100.0%
2.6.0-test3+urepl 143.1%

SDET Average Throughput (16-way P4):
2.6.0-test3 100.0%
2.6.0-test3+urepl 108.8%

The baseline is 100%, so that's an *improvement* of 43.1% or 8.8%, not 143.2/108.8.

While these are good numbers, one wonders how the "production ready" patch would change them.

(Log in to post comments)

Numbers are off

Posted Aug 21, 2003 15:52 UTC (Thu) by corbet (editor, #1) [Link]

I read the numbers right, I just wrote poorly. It was getting late in a long week... In any case, you're right; the new performance is as good as 143% of the old value, the improvement is 43%.

User data Page replication

Posted Sep 2, 2003 10:32 UTC (Tue) by balbir-_singh (guest, #887) [Link]

I think only executable text pages are replicated. As far as
I can remember data is never replicated (I worked on such systems
about two years ago).

The title can be misleading if what I am saying is correct.

Balbir