mps0-troubles

[Kenneth D. Merry]

On Fri, Feb 04, 2011 at 14:50:34 +0100, Joachim Tingvold wrote:
> On Thu, Feb 03, 2011, at 23:10:56PM GMT+01:00, Kenneth D. Merry wrote:
> >Try running this, and then do 'sysctl hw.mps' and let's see what  
> >your low
> >water mark is for free chain elements.  We'll also want to make sure  
> >your
> >chain_free value is about equal to MPS_CHAIN_FRAMES when the system is
> >idle.  On my system with a LSI 9201-16i controller, I see:
> 
> While idle;
> 
> 	[jocke at filserver ~]$ sysctl hw.mps.0
> 	hw.mps.0.debug_level: 0
> 	hw.mps.0.allow_multiple_tm_cmds: 0
> 	hw.mps.0.io_cmds_active: 0
> 	hw.mps.0.io_cmds_highwater: 53
> 	hw.mps.0.chain_free: 2048
> 	hw.mps.0.chain_free_lowwater: 2029
> 	hw.mps.0.chain_alloc_fail: 0
> 
> After I copied a 22G file from 'storage' to 'zroot';
> 
> 	[jocke at filserver ~]$ sysctl hw.mps.0
> 	hw.mps.0.debug_level: 0
> 	hw.mps.0.allow_multiple_tm_cmds: 0
> 	hw.mps.0.io_cmds_active: 0
> 	hw.mps.0.io_cmds_highwater: 99
> 	hw.mps.0.chain_free: 2048
> 	hw.mps.0.chain_free_lowwater: 2029
> 	hw.mps.0.chain_alloc_fail: 0
> 
> After I copied a 22G file from 'zroot' to 'storage';
> 
> 	[jocke at filserver ~]$ sysctl hw.mps.0
> 	hw.mps.0.debug_level: 0
> 	hw.mps.0.allow_multiple_tm_cmds: 0
> 	hw.mps.0.io_cmds_active: 0
> 	hw.mps.0.io_cmds_highwater: 897

Looks like there were a lot of commands active at once.

> 	hw.mps.0.chain_free: 2048
> 	hw.mps.0.chain_free_lowwater: 2029
> 	hw.mps.0.chain_alloc_fail: 0

But no more than 19 chain elements used at any one time.

Perhaps it could depend on memory fragmentation somewhat.  Over time you
may see the low water mark go down a bit.

The good news is that it doesn't look like we have a leak.

Ken
-- 
Kenneth Merry
ken at FreeBSD.ORG