>> On Fri, 18 Aug 2006 12:30:08 -0400, "Kauffman, Tom" <KauffmanT AT NIBCO DOT
>> COM> said:
> TSM will mount and start as many sessions as possible, and the rest will
> be in 'wait media' state until the tape they need (or a tape drive)
> becomes available. Given your scenarion, you may have one session
> reading tape and the other four waiting for access to the same tape.
> When the first session has retrieved ALL the files it needs from the
> first tape, it will go to the second (if not in use) and the session
> that has been waiting the longest will now get access to that first
> tape.
Several important points: The session that has been waiting the
longest will get access to the tape -drive-, and if the
longest-waiting session desires the mounted tape, that's all good.
But the chances of that aren't great: each restore will calclulate
its' list of desired tapes and its' desired order independantly, and
each one will walk through that order linearly. Period. (last time I
checked).
This means that if hosts A and B both want tapes T0 and T1, and they
both pick that order, and A gets T0 first, B will not fall back to
deal with T1 before it blocks on T0 access. That dynamic reordering
of work is just not something TSM does at the moment, and I can't
blame them too much. Sounds hard in most cases.
Implications for the case Troy suggested (5 simultaneous restores, 4
tapes drives, many noncollocated tapes) are that it is extremely
unlikely that any mounted tape will be cleanly passed from one restore
process to another; transitions will usually require a dismount and a
mount.
As usual, anyone got evidence that TSM's gotten smarter since last I
looked, shoot me down.
- Allen S. Rout
|