Could be that you are hitting the SATA IOPS limit. Cache does not help much
for long streaming IOs like TSM storage pools. Since SATA has only the
capacity for about 60 IOPS per drive, and the XIV has 180 drives. This is only
a maximum 10800 IOPS, At 4 KB block size this is 42 MB/sec, in the
neighborhood of what you are seeing. If you can get your block size up, the
XIV will do much better. I have tuned my system for TSM's 256 kB block size
and I have seen very satisfactory speeds in my disk pools. Just today I was
watching and the XIV storage pools were going at about 180 MB/sec, about 680
IOPS, and with 5 ms latency. This on a near full XIV with only 10% dedicated
to TSM.
Orville Lantto
-----Original Message-----
From: Richard Rhodes <rrhodes AT FIRSTENERGYCORP DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Sent: Thu, Sep 30, 2010 11:22 am
Subject: Re: [ADSM-L] [adsm] TSM and XIV
Not just 100% but very long response times. You can use iostat to see
esponse times.
Remco Post
<r.post AT PLCS DOT NL>
Sent by: "ADSM: To
Dist Stor ADSM-L AT VM.MARIST DOT EDU
Manager" cc
<[email protected]
.EDU> Subject
Re: [adsm] TSM and XIV
09/30/2010 11:24
AM
Please respond to
"ADSM: Dist Stor
Manager"
<[email protected]
.EDU>
mmm,
In that case, wouldn't the disks show 100% busy in nmon? AIX doesn't know
bout the XIV back-end processing. And, we also see these pauses on other
SM servers that are connected to an SVC.
No, I really think that something in TSM is slowing us down...
--
Gr., Remco
On 30 sep. 2010, at 11:20, Lloyd Dieter <ldieter AT ROCHESTER.RR DOT COM> wrote:
> I'm betting the pauses are the XIV destaging data from cache to physical
isk.
Not a big fan of using heavily cached disk LUNs for disk/file storage
ools....
On Wed, 29 Sep 2010 13:47:47 +0200
Remco Post <r.post AT PLCS DOT NL> wrote:
> Hi all,
>
> I'm currently testing TSM 5.5.4 on AIX 5.3 with an IBM XIV box.
>
> When I use dd or other tools to copy data off the disk to tape (LTO4),
e
> get quite a good performance, 100 MB/s or better. Even when backing up
ata
> on the XIV via shared memory directly to tape, we're quite happy, we can
> read the data at over 100 MB/s for one backup job, and for over 200 MB/s
or
> two jobs. But, when TSM uses the XIV for DISK volumes, we're not in a
appy
> place, backing up data from one XIV to TSM with diskpool on another XIV,
e
> get about 55 MB/s per backup job, best case. Both XIV boxes are
therwise
> completely idle. When migrating data of the diskpool to tape, it's the
ame,
> no matter what we do, we don't even get close to the LTO4 native
> performance, about 70 MB/s is the best I've seen, and usually it's less.
he
> TSM server is at that time only running the migration, nothing else.
>
> The stragest thing we notice is that TSM seems to completely pause every
o
> often, no disk i/o, no tape i/o no cpu utilization, nothing for about
ne
> second, and the it goes again. When I let two migration processes run,
his
> is less obvious, because one process continues while the other one
auses.
>
> We've opened a hardware call with IBM to find out if there are any
ettings
> on the hdisks or HBAs that we need to change, and even though we did get
> some hints, and some performance improvement out of that, we fell that
SM
> should be able to do a lot better.
>
> Does anyone else have experience with XIV as a diskpool? and if so, what
> sort of performance do you see?
>
> --
> Met vriendelijke groeten,
>
> Remco Post, PLCS
--
-----------------------------------------
he information contained in this message is intended only for the
ersonal and confidential use of the recipient(s) named above. If
he reader of this message is not the intended recipient or an
gent responsible for delivering it to the intended recipient, you
re hereby notified that you have received this document in error
nd that any review, dissemination, distribution, or copying of
his message is strictly prohibited. If you have received this
ommunication in error, please notify us immediately, and delete
he original message.
|
