Re: Excessive IO wait with megaraid card

We've also had significant problems with the Megaraid card on our 
Celestica Quartet quad opterons.  Most notably, RAID-5 is total crap on 
it.  Trying to run a database (MySQL in our case) on a Megaraid RAID-5 
backend was too slow to even be useable.  We ended up in a RAID 1+0 
configuration that performed quite well.

You might also try the 'megaraid2' driver instead of the 'megaraid' 
driver that is used by default.  We found it to be more stable in some 
configurations.

Paul Krizak                         5900 E. Ben White Blvd. MS 625
Advanced Micro Devices              Austin, TX  78741
Linux/Unix Systems Engineering      Phone: (512) 602-8775
Silicon Design Division             Cell:  (512) 791-0686


Jussi Silvennoinen wrote:
>> We have a (4) sun v40z's here to run some oracle databases. They have 6 
>> HD's in a hardware raid off of a LSI Logic MegaRAID 413Y (as cleaned 
>> from /proc/scsi/megaraid/1)
>>
>> Performance is fine doing it's main job (serving about 6 Lightly used 
>> Oracle databases).
>>
>> We are finding, however, that should you attempt to run a large tar job 
>> on the system, all four of the quad opteron processors end up stuck with 
>> 100% IO wait (when tarring say about a 4gb oracle database).
>>
>> We've tried using the megaraid2 driver which did improve things a bit, 
>> the system is a little more responsive during the tar, but not by much.
>> Does anyone have any ideas?
>>
>> Either how we narrow down the problem/get more info, or (ideally ;)) a 
>> fix to our rather annoying problem? It seems to appear on all four 
>> boxses, using RHEL U7 (installed with U0, but up2dated), and using 
>> redhat-made drivers.
> 
> Do you have writeback-cache enabled on the logical drives? And do find out 
> which controller you have, 413Y is just the firmware version :)
> LSI has software on their site to manage the controller from linux.
> 
> In my experience, there is something significantly wrong with Megaraid 
> when WB-cache is not enabled on the controller. I have numbers for 
> sequental writing on RHEL4, which uses a different driver though.
> 
> 320-1 2-disk raid1
>        controller WB-off write 15mb/s
>        controller WB-on write  34mb/s
>        read                    49mb/s
> 
> These numbers are simply 'time sh -c "dd if=/dev/zero of=foo bs=256k 
> count=32k ; sync" and count the transfer rate from that.
> 
> 


--
Taroon-list mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/taroon-list