ADSM-L

Re: [ADSM-L] Am I Shoe Shining My LTO4's?

2010-01-29 10:09:09
Subject: Re: [ADSM-L] Am I Shoe Shining My LTO4's?
From: "Allen S. Rout" <asr AT UFL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Fri, 29 Jan 2010 10:07:47 -0500
>> On Thu, 28 Jan 2010 14:44:35 -0600, "Hart, Charles A" <charles_hart AT UHC 
>> DOT COM> said:

> We push our 2nd copy over individual FCIP links with write acceleration
> enabled, we use to trunk 6gige links together but ran in to latency
> issues without write acceleration enabled and the technology did not
> allow for write acceleration and trunking so we end up with 3-4 LTO4
> drives down 1gige link with write acceleration.

> That said when I run select against the summary table to get timings on
> Backup stgpools the more the more processes we kick off at one the
> slower the speed (makes sense only so much per gige).  (TSM Env AIX p550
> / IBM LTO4)

> LTO4's minimum matching drive speed is 30MBS, so once we push 4
> processes down one link we are pushing less than 30MBS, thereby in
> theory running below the minimum rated drive speed we should be shining
> the drives pretty good.  Does anyone know of a way to validate that
> indeed were shining?


You can't really prove you're shoeshining from the TSM side, but you
can give yourself a pretty good guess by plotting average bandwidth in
the aggregate.  If your sum at any moment is well under (30M/s *
number of tapes used) then you pretty much know you're shining some.

Another instantaneous check is nmon (if you're on AIX) which will let
you see how much network you're using at a time.


Even if you have things bonded, keep in mind that several of the
algorithms which select links for given flows are deterministic by
measures that give TSM flows problems.

For example, however many Gb links you have, if the link selection at
the router happens deterministically with inputs of src and dest IP,
then all your traffic between TSM servers HOME-1 and REMOTE-2 will hit
the same link.

If you have enough TSM servers to be reasonably random, that will be
less of a problem, but you'll still likely detect contention
sometimes.

If links are chosen by IP and destination port you can possibly change
things up by adding TSM instances. And if your implementation is smart
enough to chose link according to the both IPs and both ports, then
you ought to actually be pretty well hashed.

But we had to fiddle with that for our remote installation, and ended
up with multiple source machines.


- Allen S. Rout