HEADS UP: Major CAM performance regression

From: Scott Long <scottl_at_samsco.org>
Date: Fri, 13 Feb 2009 03:55:53 -0700
All,

A major performance regression was introduced to the CAM subsystem in
FreeBSD 7.1.  The following configurations are known to be affected:

VMWare ESX
VMWare Fusion
     (using bt or lsilogic controller options)
HP CISS RAID
Some MPT-SAS combinations with SATA drives attached
     (Includes Dell SAS5/ir, but not PERC5/PERC6).

Pure SCSI and SAS subsystems likely are NOT affected.  Any hardware
that uses the 'ata' driver is also definitely NOT affected.  To 
determine if your installation is affected, run the following command as 
root:

camcontrol tags da0

Substitute 'da0' with another appropriate drive device number, if
needed.  Note that this ONLY AFFECTS 'da' DEVICES.  If your disks are
'ad' devices, they are NOT affected.

The result from running this command should be an output similar to the
following:

(pass0:mpt0:0:8:0): device openings: 255

If, instead, it reports a value of '1', you are likely affected.  Note
that it may be normal for USB memory devices to report a low number.
Also, many legacy SCSI disks, and devices that are not disks, may also 
be expected to report a low number.

The effect of this problem is that only one I/O command will be issued
to the controller and disk at a time, instead of overlapping multiple
commands in parallel.  This causes significantly higher latency in
servicing moderate and heavy I/O workloads, leading to very poor
performance.  Performance can be easily compared by downgrading to
FreeBSD 7.0.

I have committed a fix for this problem for FreeBSD 8-CURRENT as of SVN
revision 188570. FreeBSD 7-STABLE will be updated with the fix in a few
days once I've gotten confirmation that the fix works and doesn't cause
any adverse side-effects.  Anyone wanting to help in this validation
effort should apply the attached patch to their kernel source tree and
recompile.  Please contact me directly by email to report if the problem
is fixed for you.

If the validation process goes smoothly, I will work with the release
engineering team to turn this fix into an official errata update for
FreeBSD 7.1.

Thanks in advance for your help.

Scott




Received on Fri Feb 13 2009 - 09:55:59 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:42 UTC