We have two AIX ADSM servers with 3494ATL's & 8 3590 drives...
We also have one with 2 3590's, and another 5 with 4 3590's
then our MVS have 3490E's and a bunch of them... but who backs up
straight to tape ?
We have and continue to experiment with backing up straight to tape.
We have found what works best for us is to front end the environment
with a large diskpool. We run 250 GB to 350 GB diskpools that all
backup and archive data initially gets dumped to.
We push 30+ish GB/hr from a single AIX adsm client (in the form of a
data base archive, the client has established 9 concurrent sessions
with the single server) to an AIX adsm server now if we get out our
calculators we will find that is about what one would expect to see,
max traffic wise, over a fddi link. Now these 3590 drives are high
speed, they hate to wait on the little (or big) packets coming across
the network. Compare it to driving... if going from A to B and there
are 10,000 stop lights between them will you get there sooner if you
hit all the lights green or if you hit all the lights at RED!
Uh oh... light speed leap.
Warp on over here to bottle-necks for a second...
1) fddi 100Mb/sec (/8=12.5MB/sec, *3600=45000MB/hr, /1024=43.9GB/hr)
2) FWDiffSCSI 40MB/sec (*3600=144000MB/hr, /1024=140.6GB/hr)
and say you have 14 devices on that string... each device AT BEST
could only hope for 10 GB/hr average with all devices busy.
3) concurrent use ? Tape, no matter how fast is oriented at serving a
single task. Disk is geared towards the multi-tasking environment.
So I see below someone needs 100GB/hr backups... Well, 1st I'm an
80%'er, that is I count on at best getting for use 80% of hardware
resources for use... the rest is overhead or idle time due to limits
of other hardware... So fddi's 80% gives aroung 32GB/hr so you would
have to have 4 fddi cards out and four fddi cards in somewhere (oh,
and on different networks) for you to see 100+GB/hr moved. Plus the
CPU & other junk to keep the fddi cards supplied with data to send.
Other Junk means you would have to have at least TWO FWDiffSCSI
controllers 'cause my 80% rule says count on 112GB/hr but you are
wanting to feed 4 fddi cards we are praying will take 32GB/hr so as to
not waste resources which totals to 128GB/hr, which is more than
112GB/hr
OH and I almost forgot... all that only allows for microcode level
activity ! Any application program, system program, or user activity
would rob time from the backups!
and all this leads back to what do you reall need, what are you really
willing to spend
If I remember correctly IBM rates the 3590 drives at 31-ish GB/hr...
When I have a diskpool full of 1/2 to 1 GB archived, Oracle DB files I
see the migrations (diskpool of SSA drives to tapepool of 3590 drives
with one drive per controller) move 22GB/hr avg. (per migration
process running 2 concurrent for a migration rate of about 44 GB/hr)
So that 22GB/hr is right at my 80% rule of thumb so I'm tickled pink
;-)
later,
Dwight
______________________________ Reply Separator _________________________________
Subject: Re: How many 3590's can ADSM support?
Author: ADSM-L (ADSM-L AT VM.MARIST DOT EDU) at unix,mime
Date: 9/11/97 5:39 PM
Pls, see the IBM Storage Web page. They have a report about backing up a
huge Oracle data base (more than 500 GB) with 16 3590s simultaneously
attached to RS/6000. It's very interesting. Hope this helps.
Rafael.
At 16:00 11/09/1997 EST, you wrote:
> Hey- sounds like your talking with marketing.
>
> I've got 8 3490E's in my 3494 and haven't seen any degredation since
> the time I had only two.
>
> If your 3590's will be scsi attached, please forward all 3590
> performance numbers to the nearest junk bin. During all of our
> testing, scsi attached 3590's beat scsi 3490E by no more than 10%.
>
> No bang for the $. I was getting 4 meg/sec on 3590 vs 2.8 on 3490e.
>
> BTW the 3490/3590 test was on standalone drives, each on a different
> scsi bus, using sysback to backup raw devices.
>
> ;(
>
>
>______________________________ Reply Separator
_________________________________
>Subject: How many 3590's can ADSM support?
>Author: "Coleman; Barbara - WWH" <colemba AT WITCO DOT COM> at ~Internet
>Date: 9/11/97 12:12 PM
>
>
>I am thinking about getting 8 3590's in a 3494 library.
> Currently we are running 1 ADSM server on a wide node in an SP complex.
>
>An IBM rep. has told us that a single ADSM database can only drive up to 4
>tape units concurrently before performance degradation. They also said that
>ADSM on AIX can only support the backup of 50GB an hour.
>
>IBM has recommended that we have at least 2 ADSM servers on 2 SP nodes. Each
>ADSM server would then be connected to 4 drives in the 3494. This is not a
>realistic solution for us.
>It also doesn't seem possible that it really is supposed to work like this.
>
>Our requirements are that we back up a 200GB database in 2 hours - (backup
>at a rate of 100GB/hr.) And, we want to be able to run all 8 tape units at
>one time with only 1 ADSM server. Some tape units will be backing up other
>platforms at the same time that we are backing up the 200BG database.
>
>Does anyone have information about what ADSM will support?
>Is anyone using ADSM on AIX with more than 4 tape units? If so, what kind of
>performance are you seeing?
>Is anyone backing up 100GB/hr or more? How?
>
> -Barbara Coleman
>E-MAIL: colemba AT witco DOT com
>
>
==================================================
* *
* Rafael J. Goldschmidt *
* Elcano 5039 - (1427) Buenos Aires - Argentina *
* Telefonos: (54-1) 553 9909 / 421 4782 *
* E-mail (Internet) : rafagold AT cano.com DOT ar *
* *
==================================================
|
