I've done reasonably comprehensive testing of this myself over the years. One
thing I noticed lately is the effect of changing movebatchsize from the default
500 to the maximum 1000. Almost doubled the performance I saw during backup
stg operations (I use backup stg so I can repeat the test over and over: delete
the copy storage pool volume and do the test again...)
Here are two four stream backup stg tests I recently ran. I'll describe each
1. IBM x3800 Win2003 Standard TSM 5.3.4.1, 8*300GB 10K drives using DISK device
class cachepool. Qualstar XLS library, 4 LTO3 DFA 4Gb/sec Switched.
Test Total Data (MB) Total Files Average Size Begin
Time End Time Elapsed Time (Seconds) Data Transfer Rate
(MB/sec)
Agregate Processes 532,672 345,052 1.54 9:49:51
10:41:18 0:51:27 3,087 172.55
Process 1 133,168 86,263 1.54 9:49:51
10:39:53 0:50:02 3,002 44.36
Process 2 133,168 86,263 1.54 9:49:51
10:36:33 0:46:42 2,802 47.53
Process 3 133,168 86,263 1.54 9:49:51
10:41:18 0:51:27 3,087 43.14
Process 4 133,168 86,263 1.54 9:49:51
10:40:16 0:50:25 3,025 44.02
I thought this was pretty good performance and believe that the bottleneck in
this case is probably the disk pool.
2. HP ML370G4 Win 2003 Standard TSM 5.3.4.1, Xyratex 4Gb/sec 12*500 SATA,
RAID50, three Logical Drives of 1.65 TB each. File device class. Spectralogic
T120 4*LTO3, SCSI Attached.
File device class on the Xyratex backed up to four LTO3 drives SCSI connect
Test Total Data (MB) Total Files Average Size Begin
Time End Time Elapsed Time (Seconds) Data Transfer Rate
(MB/sec)
Aggregate Processes 530,202 286,426 1.85
13:53:07 14:49:57 0:56:50 3,417 155.17
Process 1 127,882 117,077 1.09
13:53:07 14:49:57 0:56:50 3,417 37.43
Process 2 139,948 14,604 9.58
13:53:07 14:49:24 0:56:17 3,377 41.44
Process 3 136,259 51,729 2.63
13:53:07 14:48:28 0:55:21 3,321 41.03
Process 4 126,113 103,016 1.22
13:53:08 14:48:22 0:55:14 3,314 38.05
Kelly J. Lipp
VP Manufacturing & CTO
STORServer, Inc.
485-B Elkton Drive
Colorado Springs, CO 80907
719-266-8777
lipp AT storserver DOT com
-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of
Henrik Wahlstedt
Sent: Tuesday, January 09, 2007 8:27 AM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Re: [ADSM-L] Performance with move data and LTO3
Hello,
A late answer about LTO3 performance, but for the records. Origal post is below.
I used 'audit vol volume_name skippartial=y fix=n' and 'move data volume_name
reconstr=no'.
Since the tape drives are not connected to a switch I used the information and
timing from actlog when the tape is opened as an input volume and when the
processing is done. Maybe not the best way but it will give me a clue about the
performance.
With IBM drives on Windows:
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='360023'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
360023 762938.0 34.2 0.0
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='360024'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
360024 762938.0 49.6 0.0
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='360125'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
360125 762938.0 45.6 0.0
360023
Audit: 196306 items / 33m04sec
Move data: 196306 items / 273,902,675,481 bytes / 65m22sec ~69,84Mb/s
360024
Audit: 93202 items / 41m57sec
Move data: 93202 items / 397,207,010,912 bytes / 53m29sec ~123,8Mb/s
360125
Audit: 21470 items / 71m02sec
Move data: 21470 items / 369,229,158,737 bytes / 104m28sec ~58,9Mb/s
With HP drives on Linux:
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='350075'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
350075 409600.0 41.6 0.0
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='350204'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
350204 441710.3 68.9 31.1
select volume_name, est_capacity_mb, pct_utilized, pct_reclaim from volumes
where volume_name='350257'
VOLUME_NAME EST_CAPACITY_MB PCT_UTILIZED PCT_RECLAIM
------------------ -------------------- ------------ -----------
350257 463674.8 49.5 50.6
350075
Audit: 705210 items / 38m30sec
Move Data: 705210 items / 179,065,879,072 bytes / 36m48sec ~81,1Mb/s
350204
Audit: 36603 items / 71m35sec
Move Data: 36603 items / 319,319,489,218 bytes / 75m11sec ~70,79mb/s
350257
Audit: 95345 items / 54m06sec
Move Data: 95345 items / 240,739,702,948 bytes / 46m26sec ~86,4Mb/s
I guess Wanda is right about the buffer usage on Windows when performing a Move
data. Audit seems OK. One thing is for sure, as always, backup stgpool will be
disk to tape...
On the other end I´m curious about the Linux system that performs Move data
faster than Audit. And have less variance in speed than the Windowsbox.
However my main interest was performance of LTO drives and (HP vs IBM drives).
I guess our Linux box will outperform the Windows box with adequate hardware.
//Henrik
-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of
Salak Juraj
Sent: den 22 december 2006 11:44
To: ADSM-L AT VM.MARIST DOT EDU
Subject: AW: Performance with move data and LTO3
Hi!
I saw some related information (source:Quantum) under
http://www.datastor.co.nz/Datastor/Promotions.nsf/4a91ca5e06d20e15cc256ebe0002290e/d954d1c5e5e6df09cc25723b00740956/$FILE/When%20to%20Choose%20LTO3%20Tape%20Drives.pdf
best
Juraj
> -----Ursprüngliche Nachricht-----
> Von: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] Im Auftrag
> von Henrik Wahlstedt
> Gesendet: Donnerstag, 21. Dezember 2006 15:43
> An: ADSM-L AT VM.MARIST DOT EDU
> Betreff: Re: Performance with move data and LTO3
>
> Nice one! I get back on this after the Holidays.
>
> Thanks
> Henrik
>
> -----Original Message-----
> From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf
> Of Prather, Wanda
> Sent: den 19 december 2006 18:15
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Re: Performance with move data and LTO3
>
> Interesting.
>
> Differences in IBM vs. HP LTO3 drives:
>
> I have been told that the IBM drives do "smart"
> compression using a bypass buffer. If a block of data is going to
> expand during compression, the IBM drives will stop compression and
> write the uncompressed block, which should make them a bit faster.
>
> Re tape to tape operations:
>
> I have observed the same behavior; tape to tape operations are
> inexplicably slower than you would expect when the TSM server is on
> WINDOWS. I have observed this with fibre drives, and SCSI drives,
> 3590 and LTO. I suspect it has something to do with buffer use, but
> since Windows provides no tools whatever to measure performance of
> tape devices or buses with tapes on them, I've never been able to make
> any other determination.
>
> I don't think it is a READ issue with the drives. Try testing using
> an AUDIT; that just reads the tape and doesn't write anything. I
> suspect you'll get faster READ times. I would be interested in seeing
> your results!
>
>
>
> Wanda Prather
> "I/O, I/O, It's all about I/O" -(me)
>
>
>
>
> -----Original Message-----
> From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf
> Of Henrik Wahlstedt
> Sent: Thursday, December 14, 2006 10:37 AM
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Performance with move data and LTO3
>
> Hi,
>
> I wonder what transfer rates (move data from drive to drive) I am
> supposed to see with LTO3.
>
> I have two TSM servers, one 32-bit Win2k3 and one 64-bit
> 2.6.9-11.Elsmp, with a SL500 and FC LTO3 drives.
> Similar HW (HP DL585) except for one server have HP- and the other
> have IBM drives. Drives are on separate PCI busses.
>
> I used a dataset of 50Gb with large files, same file type on both
> systems. Only scratch tapes and no expiration on the datasets.
> No other tape activity on the systems during the tests.
>
> I tested disk->mt0->mt1->mt2->mt3->mt1->mt0->disk
> From disk to tape I get a throughput of 74-76Mb/s with IBM drives,
> (migration).
> From tape to tape, (move data), with HP drives I get a throughput of
> 30-46Mb/s and with IBM drives I get 39-59Mb/s.
> From disk to tape, (move data), with IBM drives I get a throughput of
> 44Mb/s.
>
> Apperently write speed seems OK but read spead is an issue?!
> Or is this normal?
>
>
>
> Thanks
> Henrik
>
>
>
>
>
>
>
>
>
>
>
>
>
>
-------------------------------------------------------------------
The information contained in this message may be CONFIDENTIAL and is
intended for the addressee only. Any unauthorised use, dissemination of the
information or copying of this message is prohibited. If you are not the
addressee, please notify the sender immediately by return e-mail and delete
this message.
Thank you.
|
