From: John Hawkes <hawkes@babylon.engr.sgi.com>
To: linux-kernel@vger.kernel.org
Subject: Re: [BENCHMARK] I/O regression after 2.6.0-test5
Date: Mon, 27 Oct 2003 15:33:52 -0800 (PST) [thread overview]
Message-ID: <200310272333.h9RNXqaP2542634@babylon.engr.sgi.com> (raw)
I ran AIM7 on a 64p Altix NUMA (Itanium 2 Madison CPUs, 1.5 GHz), with an
XFS filesystem on each of 99 fiberchannel disks, with two different flavors
of AIM7 workloads, on both 2.6.0-test5 and 2.6.0-test8. One workload shows
that -test8 is half the throughput of -test5. The other workload shows
that -test8 is 22% faster.
The first workload is the default AIM7 "multiuser/shared system" workload.
2.6.0-test8 shows a peak throughput of 52% of -test5 -- it drops in half.
The -test8 peak at 64p is roughly the same as at 8p and 4p. This
bottleneck cannot be attributed solely to I/O hardware, since our SGI
ProPack kernel (based upon 2.4.21) achieves a peak throughput at 64p that
is 30% higher than 2.6.0-test5's peak at 64p.
The second workload takes the same "multiuser/shared system" workload,
but turns off the O_SYNC flag for the three sync_* subtests. This
generally produces higher peak throughput values than when using the
default O_SYNC, and the peak throughput values scale significantly better
as CPU counts increase. Here the -test8 kernel is 22% *faster* than
-test5 at 64p (and 8x faster for no-O_SYNC than with-O_SYNC).
John Hawkes
next reply other threads:[~2003-10-27 23:33 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2003-10-27 23:33 John Hawkes [this message]
-- strict thread matches above, loose matches on Subject: below --
2003-10-29 0:38 [BENCHMARK] I/O regression after 2.6.0-test5 rwhron
2003-10-26 10:38 rwhron
2003-10-27 23:49 ` Dave Olien
2003-10-24 23:43 rwhron
2003-10-24 12:46 rwhron
2003-10-24 0:10 rwhron
2003-10-21 13:05 rwhron
2003-10-21 15:57 ` Nick Piggin
2003-10-21 21:07 ` venom
2003-10-22 9:54 ` Nick Piggin
2003-10-22 18:30 ` Dave Olien
2003-10-22 19:08 ` Dave Olien
2003-10-23 2:24 ` Nick Piggin
2003-10-23 3:48 ` Nick Piggin
2003-10-23 20:35 ` Dave Olien
2003-10-23 23:07 ` Nick Piggin
2003-10-20 0:37 rwhron
2003-10-20 1:35 ` Nick Piggin
2003-10-20 4:52 ` Andrew Morton
2003-10-20 8:12 ` Nick Piggin
2003-10-20 4:51 ` Dave Olien
2003-10-20 7:55 ` venom
2003-10-21 4:49 ` Nick Piggin
2003-10-24 13:52 ` Magnus Andersson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=200310272333.h9RNXqaP2542634@babylon.engr.sgi.com \
--to=hawkes@babylon.engr.sgi.com \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).