Jim,
few hints:
- TSM restores and reclamation for big aggregates are read/seek
forward which is not actually start/stop so I think they should not count,
but ... restores and reclamation of small files from non-collocated
volumes are definitely start/stop intensive. How intensive are for you
depends on how "semi"-collocated your tapes are. TSM aggregation mitigates
starts/stops but cannot fully relieve them. With such high number of
clients mounts/dismounts prevale starts/stops and you'd better stay with
3590.
- some of the servers with big files/databases can go to a stgpool
on 3584. LTO shines with big files and is still very good with files over
few MB (thanks to TSM's aggregation). Filesystems where aggregation can
get 50-100 MB for every TXNGROUPMAX files are just fine.
- reaching 5.1.6 on your forthcoming upgrade you can upgrade to
3590H. This would give you 50% more - 1700x10 GB (uncompressed) keeping
fast seeks/restores right now. Usage of K cartridges will also increase
the capacity without performance degradation. There is no problem to mix
J-s and K-s in the library. And for the future - the 1 TB cartridge
announced by IBM will definitely go to strengthen 3590 line and I guess
this would happen during this year.
So the answer to the subject - LTO can't replace 3590! Upgrade slowly and
it would be fine.
Zlatko Krastev
IT Consultant
James R Owen <Jim.Owen AT YALE DOT EDU>
Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
30.01.2003 21:36
Please respond to "ADSM: Dist Stor Manager"
To: ADSM-L AT VM.MARIST DOT EDU
cc:
Subject: ?Can LTO replace 3590E tape for busy TSM backup/restore
service?
??? Do you have experience w/ or plans for replacing 3590E with LTO
tapes
for a busy enterprise TSM backup service? Is LTO performance OK
???
We expect to double business for one of our TSM services over next 4
years.
Last year, for space and cost savings, future enhancements, etc...
we bought new LTO (100GB/tape) libraries to use for Copy STGpools,
replacing existing 3590E (20GB/tape) Copy STGpools which made all of
those 3590E tapes available to expand our primary STGpools.)
Now, we are considering whether we can use LTO tapes for new primary tape
STGpools or should we continue to expand our existing 3590E tape STGpools?
We have performance concerns about using LTO tapes for primary STGpools!
Our current environment:
-----------------------
TSM v4.2.3.2+ (upgrading to v5... soon, before 5/2003)
AIX v4.3.3 (upgrading to v5... soon, before 5/2003)
IBM 7025-6F1 w/ 2*600Mhz CPUs and 1GB memory
IBM 3494 w/ 6 3590E drives for 1700 primary backup tapes
IBM 3584 w/ 4 3580 LTO drives for 300 copy + 10 TSM DB backup
tapes
2700 active clients (>100 Servers, >2500 Win*,Mac,Linux,etc. PC's)
backup 5-600GB/night (some 1-4GB files, mostly much smaller
files!)
TSM DB: 74-76% of 120GB
TSM LOG: 12GB (w/ LOGM=R needs 1 full + 1-2 incr.DB bkups daily)
Our expectations for next 4 years:
---------------------------------
clients: +25%/year, @+4 years = +2800 clients: +100 S, +2700 PC's
backups: +25->50%/year, @+4 years = 2-4 * current backup load!
how: ??? replicate another or "super-size" current TSM service?
Our current LTO tape experience:
-------------------------------
We use LTO tapes ONLY for an online Copy STGpool and TSM DB
backups.
LTO seems to be ~100% faster than 3590E tape for TSM Backup DB...
LTO can be 30-50% slower than 3590E tape for TSM BAckup STGpool...
w/ EMC SAN disks or 3590E tapes -> LTO vs 3590E Copy
tapes.
LTO can be 30-50% slower than 3590E tape for our online Copy
STGpool
reclaiming LTO Copy tape -> LTO Copy tape vs 3590E ->
3590E.
Do you see similar or better performance?
----------------------------------------
Our concerns about replacing 3590E w/ LTO tapes for primary STGpools:
--------------------------------------------------------------------
We are worried about start/stop and seeking performance using large
capacity
LTO tapes for restoration or reclamation of aggregated small/medium-sized
files on "semi-collocated" tapes. [We use collocation, but currently have
350/1700 "filling" 3590E tapes available for 2700 clients, so we often
have
multiple clients' backups sharing the same 3590E tape. This problem could
be much worse with 100GB LTO tapes as we might have only 70/320 "filling"
LTO tapes.] Even w/ full-collocation, we believe restoration/reclamation
of a highly active client's aggregated backups of many small/medium-sized
files might involve substantial start/stop and seeking operations and
would
probably perform poorly on LTO when compared to 3590E tapes.
Are our concerns warranted?
--------------------------
In IBM's 2003-01-28 announcement of new Ultrium 2 LTO technology,
the last paragraph under the heading "Product Positioning" states:
For mission-critical data protection, optimized for
enterprise multi-mode and host attachment, high-cycle
and start/stop intensive tape applications, consider the
proven IBM TotalStorage Enterprise Tape Drive 3590 or
the IBM TotalStorage Enterprise Automated Tape Library
3494.
Are TSM restores and reclamation of aggregated files start/stop intensive?
Has anyone done TSM performance comparisons of LTO vs 3590E for:
---------------------------------------------------------------
- DR or FS restore from (semi-collocated) tapes?
- primary (semi-collocated) tape->tape reclamation?
Thank you for any experienced advice you will share with us.
--
Jim.Owen AT Yale DOT Edu (203.432.6693)
|
