Andy,
I understand that given TSM's current functionality the source code
probably runs 10s of millions of lines. And I certainly don't expect
anyone to read or memorize that.
But there are some common operating techniques built into TSM that haven't
changed much over the years. Round-robin'ing drives within a library is
one example. Volume sequencing is another.
Was this info "critical?" No. But it would certainly be useful. In
this case I have a 40-slot production library of which I can devote 1/2 to
the export job. The other half is required to support routine operations.
I know from the volume usage that my export will read 75 tapes. So being
able to preload the first 20 tapes or so confident that I had the right
tapes meant the export would go more efficiently and I would rely less
upon the current 26 hour mount wait.
>From the forum I expected an answer only if anyone else had actually gone
through that, since it isn't documented anywhere.
I wasn't expecting IBM SW Support to take three hours to send me the long
version "haven't a clue."
In the meantime, using Excel and my ODBC connection, I created three
lists:
1) the volume usage of the node in volume label order;
Then, since volumeusage does not include any volume-specific properties, I
got the other two lists from the volumes table:
2) volume list sorted by Last Write Date;
3) volume list sorted by quantity of data, as Est_cap x Pct_util.
I figured that the first tape requested would be at one end or the other
of one of those three lists, and I was right. The first tape requested
was the oldest tape on list #2. By the time it had asked for seven tapes,
the pattern was clear. I then preloaded the next fourteen tapes
comprising the next few batches from the volume data, and allowed it to
run overnight.
I'm writing this on Saturday because I came in expecting to swap several
batches of source tapes but found the process aborted due to a jammed tape
in the output drive. But that's another issue.
The jam has been fixed, the export process restarted, and I *know* that
the first 18 tapes needed by the export process are already in the
library.
Thanks for all your help.
Tab Trepagnier
TSM Administrator
Laitram LLC
Andrew Raibeck <storman AT US.IBM DOT COM>
Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
07/25/2003 02:46 PM
Please respond to "ADSM: Dist Stor Manager"
To: ADSM-L AT VM.MARIST DOT EDU
cc:
Subject: Re: Export order
Tab,
For better or worse, this is an undocumented facet of TSM operation (just
like we don't document how we pick the management class to which
directories are bound if more than one class happens to have the same
longest RETONLY setting). It's not that *IBM* (the company) can't tell you
the answer... it's just a matter of getting to the right person who might
happen to know this bit of information. And even then, some of the code
has been written so long ago, someone would have to go back through the
code to refresh their memory.
That said, your question isn't unreasonable. If this is critical
information for you to have (please forgive me, but I forget the history
behind this thread), then they need to move this up the support chain
until they get to server development, which is where the buck stops.
Regards,
Andy
Andy Raibeck
IBM Software Group
Tivoli Storage Manager Client Development
Internal Notes e-mail: Andrew Raibeck/Tucson/IBM@IBMUS
Internet e-mail: storman AT us.ibm DOT com
The only dumb question is the one that goes unasked.
The command line is your friend.
"Good enough" is the enemy of excellence.
Tab Trepagnier <Tab.Trepagnier AT LAITRAM DOT COM>
Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
07/25/2003 12:13
Please respond to "ADSM: Dist Stor Manager"
To: ADSM-L AT VM.MARIST DOT EDU
cc:
Subject: Re: Export order
Wanda,
OK. Based on the first seven tapes out of 75, it appears that TSM asks
for source tapes in order that they were FIRST WRITTEN, from old to new.
Since there is no queryable field that reports first write date, if tapes
are used in batches, as these were, then the LAST write date is a suitable
substitute and will at least identify the order of the batches.
In other words, in my case each batch of tapes consists of about four
tapes. I know the order of the batches, but I don't necessarily know the
order within each batch. So as long as I check in complete batches, I
ensure that the export will process all files from that batch without ever
asking for a tape from outside the library.
The tape currently being read is from the June 2001 batch. All of the
tapes from that batch are already checked into the library.
If I check in all four tapes from the July 2001 batch, TSM will be able to
continue when it finishes with June, even though I don't know which July
tape will be loaded first.
Note that I opened a PMR on this yesterday and IBM couldn't tell me the
answer either!
Random indeed.
Tab Trepagnier
TSM Administrator
Laitram LLC
"Prather, Wanda" <Wanda.Prather AT JHUAPL DOT EDU>
Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
07/24/2003 01:38 PM
Please respond to "ADSM: Dist Stor Manager"
To: ADSM-L AT VM.MARIST DOT EDU
cc:
Subject: Re: Export order
I can't answer your actual question, and that's an ugly problem.
If you find the right answer, let me know!
My workaround:
Leave the tapes out of the library.
Start the EXPORT. On my SCSI library, with the tapes checked out, I get
the
"cartridge blahblah required for use, checkin within 60 minutes" message.
One solution is to just change your MOUNTWAIT time to 1440 (24 hrs) for
the
devclass, so the mount will sit and wait until you get in to morrow
morning.
But to do what you actually asked for:
CANCEL the EXPORT.
Mark that tape as DESTROYED.
Start the EXPORT again.
On my 4.2 system, an EXPORT skips the destroyed tape and calls for the 2nd
tape.
Repeat 20 times.
Mark all the tapes back to READONLY and check them all in.
Wanda
-----Original Message-----
From: Tab Trepagnier [mailto:Tab.Trepagnier AT LAITRAM DOT COM]
Sent: Wednesday, July 23, 2003 4:24 PM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Export order
TSM 5.1.7.0 Server on AIX 4.3.3
Admin client 5.1.6.0 both on Win32 and AIX
Client data originally written by IRIX client 3.1.0.7
I am preparing to export permanent archives from an old IRIX node to DLT.
The source data is spread across 75 DLTs in a primary storage pool. Those
tapes contain the contents of one filespace, but were written across a
four year period. All tapes are out of the library - they're in a box in
my office.
I will only be able to load the source library with about 20 tapes at a
time.
A preview export is not practical because poor archived directory handling
by old ADSM versions and ADSM clients that followed circular links have
generated *billions* of objects in that filespace. I discovered that last
week when I ran Expire Inventory with Skipdir=NO. I killed the process
after 3 days during which it had examined 1/2 Billion objects for this one
node/filespace alone, deleting approx 1 in 1500.
What I need is the expected sequence that TSM will load the source tapes
to obtain files for export.
Is it by last written date? If yes, then new to old, or old to new?
Is it by volume name?
Is it by quantity of data? If yes, high to low, or low to high?
Etc.
In other words, I need to be able to load about 20 tapes in the source
library confident that they will supply all required data overnight so I
can load the correct next set the following day, until all 75 tapes have
been processed.
Thanks.
Tab Trepagnier
TSM Administrator
Laitram LLC
|
