Public bug reported:
Binary package hint: linux-source-2.6.20
Using bonnie++, disk write performance is up to 50% slower on Ubuntu
7.04 (kernel 2.6.20) compared with Ubuntu 6.10 (kernel 2.6.17) and
Ubuntu 6.06 (kernel 2.6.15).
Ubuntu 7.04, 2.6.20-15-server, i386, no load
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
ziggy 32G 85938 33 53181 19 170511 29 426.2 0
Ubuntu 6.10, 2.6.17-11-server, i386, no load
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
ziggy 32G 117204 41 55691 19 174137 29 460.6
1
These tests were conducted on a dual Opteron 246 (2.0 GHz) with 16GB
ram, and a 3ware RAID controller with 10 disks in RAID-10 configuration.
64 bit kernels show similar results. Test was repeated on various other
SMP boxes, including dual dual-core Opteron and dual quad-core Xeon
systems, all with same outcome - Feisty is slow.
Problem is severely aggravated by large number of context switches, eg,
running VMware server on the box. VMware guests get disk IO timeouts,
generally run slow, and in severe cases report SCSI bus resets. All the
while the host CPUs are almost idle. If the host has several guests
running on it, and some moderate host disk IO is generated (such as
copying biggish files), the guests become unusable.
These problems do not show up if Edgy or Dapper is run on that same box.
This is not unique to just one system, and can be reliably reproduced on
a variety of SMP systems. I emphasise the SMP because I have not been
able to show any significant performance difference on a uniprocessor
Pentium D, but this may just be that the box does not have fast enough
disk IO to demonstrate the issue.
This may also be an issue with the 3ware driver, but I doubt it, since
it hasn't changed a lot in recent months. It seems more like something
is really haywire in the process scheduler.
** Affects: linux-source-2.6.20 (Ubuntu)
Importance: Undecided
Status: Unconfirmed
--
kernel 2.6.20 disk write performance much slower than 2.6.17
https://bugs.launchpad.net/bugs/113532
You received this bug notification because you are a member of Ubuntu
Bugs, which is the bug contact for Ubuntu.
--
ubuntu-bugs mailing list
ubuntu-bugs@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
