Re: bonnie++ uninterruptible under heavy I/O load

[Denis Vlasenko]

On Friday 11 March 2005 13:08, Simone Piunno wrote:
> 
> Hello,
> 
> I'm testing a pair of new servers we just bought.
> They are HP DL585 dual Opteron 844 with 8G RAM, RAID1 over 2x72G SCSI disks 
> (HP CISS driver) and running 2.6.11.  In /proc/driver/cciss/cciss0 we have:
> 
> cciss0: HP Smart Array 5i Controller
> Board ID: 0x40800e11
> Firmware Version: 2.56
> IRQ: 18
> Logical drives: 1
> Current Q depth: 0
> Current # commands on controller: 0
> Max Q depth since init: 6
> Max # commands on controller since init: 190
> Max SG entries since init: 31
> cciss/c0d0:       72.83GB       RAID 1(1+0)
> 
> As a test, I'm running bonnie++ while two md5sum are checksumming /dev/zero.
> I see 98% CPU is correctly allocated on the two md5sum processes, but when 
> bonnie is in the "Writing intelligently phase" the system appears noticeably 
> unresponsive (high latency) and bonnie and pdflush are uninterruptible almost 
> all the time (D in top output).  If I try to kill -9 bonnie++, it goes on 
> consuming I/O bandwith for several tens of seconds, before finally dieing.
> 
> Unresponsiveness is not 2.6.11 specific (we've seen the same thing on 2.6.10 
> and 2.6.8), not I/O scheduler specific ("as" and "deadline" behave the same)
> and not CPU/SMP specific (reproduced on single P4 HT and single P3), but only 
> on these two DL585 servers we've seen bonnie++ resisting kill -9 for tens of 
> seconds.
> 
> Of course on request I can provide any other useful info.
> Any help is appreciated.

I think Alt-SysRq-T will be interesting to see
--
vda

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/