Re: Poor performance with SVM mirrored stripes

From: C. Tate Baumrucker (tate@baumrucker.org)
Date: Thu Jul 13 2006 - 22:04:30 EDT


Kris, thanks for the quick reply. I'm not sure why the start block is
different, but I've been unable to overcome that despite copying the vtoc
between disks, etc.
Regardless, I found the source of the trouble and it has nothing to do
with the constructed metadevices or their svm configuration. I was
mounting the mirrored stripes as an Oracle filesystem and therefore using
the forcedirectio option, which disables the caching aspects of the
filesystem. Mounting without this option resulted in optimized and
to-be-expected I/O to the mirrored stripes.

Now, I wonder what performance impact that will have on Oracle.
Sun/Oracle claim it increases performance similar to that of a raw
filesystem deployment, but my (albeit limited) testing shows otherwise.
Once the database if fully built and data imported, I'll try some query
tests against the DB w/ and w/o forcedirectio in place.

Anyone have any thoughts?

Thanks,
Tate

Kristopher Briscoe wrote:
> right off the bat what jumps out at me is the fact that each stripe has a
> different starting block offset on each disk. This will seriously degrade
> performance.
>
> kris--
>
> On 7/13/06, C. Tate Baumrucker <tate@baumrucker.org> wrote:
>>
>> Problem: poor write performance with configuration included below.
>> Before attaching striped subdevices to mirror, they write/read very fast
>> (>20,000 kw/s according to iostat), but after mirror attachment and sync
>> performance is <2,000 kw/s. I've got standard SVM-mirrored root disks
>> that perform at > 20,000 kw/s, so something seems odd.
>> My I/O tests are done using large file transfer and bonnie++ while
>> watching iostat output.
>> Any help would be greatly appreciated.
>> Thanks,
>> Tate
>>
>> Config follows:
>> Running 5.9 Generic_117171-17 on sun4u sparc SUNW,Sun-Fire-V890.
>>
>> Configuration is 4 72GB 10K disks with the following layout:
>> d0: Mirror
>> Submirror 0: d3
>> State: Okay
>> Submirror 1: d2
>> State: Okay
>> Pass: 1
>> Read option: roundrobin (default)
>> Write option: parallel (default)
>> Size: 286607040 blocks (136 GB)
>>
>> d3: Submirror of d0
>> State: Okay
>> Size: 286607040 blocks (136 GB)
>> Stripe 0: (interlace: 4096 blocks)
>> Device Start Block Dbase State Reloc Hot Spare
>> c1t4d0s0 0 No Okay Yes
>> c1t5d0s0 10176 No Okay Yes
>>
>>
>> d2: Submirror of d0
>> State: Okay
>> Size: 286607040 blocks (136 GB)
>> Stripe 0: (interlace: 4096 blocks)
>> Device Start Block Dbase State Reloc Hot Spare
>> c1t1d0s0 0 No Okay Yes
>> c1t2d0s0 10176 No Okay Yes
>> _______________________________________________
>> sunmanagers mailing list
>> sunmanagers@sunmanagers.org
>> http://www.sunmanagers.org/mailman/listinfo/sunmanagers
>>
>
>
>
> --
> Kristopher T. Briscoe
> Sr. UNIX Engineer
> 404.353.3223
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers



This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:40:23 EDT