Does anybody in ADSM development have an authoritative answer to this question,
regarding partitioning a simple (ie, non-3494/5) tape library?
psz1 AT cornell DOT edu on 04/23/98 02:50:14 PM
Please respond to psz1 AT cornell DOT edu
To: Lindsay Morris/Lexington/Contr/IBM@ibmus
cc:
Subject: Re: partitioning robot
It really depends on whether ADSM throws atomic commands at the library, or
if it throws sequences of commands at the library. If sequences, and if
ADSM doesn't realize that LIBRARY1 and LIBRARY2 are really the same device,
then things will get confused. Let me illustrate by example, tho' I'll
start by saying that I have no idea what SCSI commands exist for the library:
ADSM commands to LIBRARY1:
POSITION TO ELEMENT 34 (a tape slot)
PULL TAPE
POSITION TO ELEMENT 117 (a tape drive)
PUSH TAPE
Now, if you've got a similar set of commands being issued to LIBRARY2 at
the same time, then there is a danger that the wrong tape will get put into
the wrong tape drive, unless there is a semaphore type of lock before and
after the sequence.
If the commands are simpler, eg:
MOVE FROM 34 TO 117
Then there wouldn't be this window of vulnerability.
..Paul
--
At 02:16 PM 4/23/98 -0400, Lindsay Morris wrote:
At 02:16 PM 4/23/98 -0400, Lindsay Morris wrote:
>I think (could be wrong) that most tape libraries just take a bunch of
COMMANDs
>on the device ADSM uses - commands like mount/dismount a tape, read/write a
>tape on a drive, audit library, etc.
>The DATA path is a different device - ie, the device of that DRIVE.
>
>So I don't think any locking is necessary. ADSM just throws commands at the
>library - those commands cna be intermixed.
>
>
>psz1 AT cornell DOT edu on 04/23/98 11:17:36 AM
>Please respond to psz1 AT cornell DOT edu
>To: Lindsay Morris/Lexington/Contr/IBM@ibmus
>cc:
>Subject: Re: partitioning robot
>
>
>Hmm.. You may be right about that. So the remaining question, as you say,
>is whether you can define 2 libraries and point them at the same device.
>I.e., does ADSM have locking support for this type of use of the library
>device.
>
>Thanks for you help on this.
>..Paul
>
>At 09:39 AM 4/23/98 -0400, Lindsay Morris wrote:
>>Actually, the only question is: can you define two ADSM libraries, point
them
>>both to the same device (/dev/libwhatever), and have them work?
>>
>>If you can do that (and I bet you can), then you can define drives, checkin
>>tapes as I discussed below.
>>
>>The only internal partitioning support the 3494/5 has is the ability to
>>categorize tapes, so you physically cannot check in somebody else's tape to
>>your library. But that simply prevents operator error. If you are simply
>>careful to check in tapes 100-200 in library 1 (say), and tapes 201-300 to
>>library 2, and never do checkouts/checkins to the wrong library, then you
>>should be OK.
>>
>>You can't share tape DRIVES - you'll need to have two drives per partition
>(or
>>else you can't do reclamation).
>>
>>
>>psz1 AT cornell DOT edu on 04/22/98 06:08:41 PM
>>Please respond to psz1 AT cornell DOT edu
>>To: Lindsay Morris/Lexington/Contr/IBM@ibmus
>>cc:
>>Subject: Re: partitioning robot
>>
>>
>>Ok, I understand you. However, the IBM Tape Library (e.g., 3494) actually
>>has support within it for partitioning itself between multiple servers.
>>The DLT library that I have does not have this capability. I think this
>>would not work for my library.
>>
>>Thanks again.
>>..Paul
>>--
>>At 03:54 PM 4/22/98 -0400, Lindsay Morris wrote:
>>>I'm not in ADSM support, but I support ADSM for IBM Global Services.
>>>Publication SG24-4409-02, "Guide to Sharing and Partitioning IBM Tape
>Library
>>>Dataservers", discusses the scheme.
>>>We have a couple of servers which use one dsmserv process to manage
>different
>>>library partitions, as I discussed below.
>>>
>>>True, if you run "AUDIT LIBRARY", the whole library gets audited, not just
>>>your partition, since the library does the work through its own computers,
>>>rather than ADSM doing the work. Kind of a bad thing, but how often do
>>you do
>>>this? And the economies of scale you get from sharing a large library
>can be
>>>considerable.
>>>
>>>
>>>
>>>psz1 AT cornell DOT edu on 04/22/98 02:11:49 PM
>>>Please respond to psz1 AT cornell DOT edu
>>>To: Lindsay Morris/Lexington/Contr/IBM@ibmus
>>>cc:
>>>Subject: Re: partitioning robot
>>>
>>>
>>>Hi Lindsay,
>>>
>>>Are you in ADSM support? Can you tell me if ADSM supports your suggestion?
>>>Things like AUDIT LIBRARY would have a problem, no? What else?
>>>
>>>Thanks.
>>>..Paul
>>>
>>>At 09:15 AM 4/20/98 -0400, you wrote:
>>>>Define two different ADSM libraries using the same /dev/whatever.
>>>>Define drive1,2 on one library, drive3,4 on the other.
>>>>Checkin one set of tapes to one library, the other set oftapes to the
>other.
>>>>
>>>>Yes?
>>>>
>>>>
>>>>ADSM-L AT VM.MARIST DOT EDU on 04/18/98 07:57:15 PM
>>>>Please respond to ADSM-L AT VM.MARIST DOT EDU
>>>>To: ADSM-L AT VM.MARIST DOT EDU
>>>>cc:
>>>>Subject: partitioning robot
>>>>
>>>>
>>>>Can anyone think of a way to partition a 4-drive DLT tape robot so that
one
>>>>set of tapes always gets loaded on 2 specific drives (e.g., DRIVE1 and
>>>>DRIVE2), but a different set of tapes always gets loaded in the other 2
>>>>drives (e.g., DRIVE3 and DRIVE4)? One set of tapes would be newly added.
>>>>The other set are pre-existing tapes containing user data.
>>>>
>>>>The robot is a Breece Hill Q47, the drives are DLT 4000s.
>>>>
>>>>Thanks for any ideas.
>>>>..Paul
>>>>
>>>
>>>--
>>>Paul Zarnowski Ph: 607-255-4757
>>>747 Rhodes Hall, Cornell University Fx: 607-254-8888
>>>Ithaca, NY 14853-2602 Em: psz1 AT cornell DOT edu
>>>
>>>
>>>
>>>
>>
>>--
>>Paul Zarnowski Ph: 607-255-4757
>>747 Rhodes Hall, Cornell University Fx: 607-254-8888
>>Ithaca, NY 14853-2602 Em: psz1 AT cornell DOT edu
>>
>>
>>
>>
>
>--
>Paul Zarnowski Ph: 607-255-4757
>747 Rhodes Hall, Cornell University Fx: 607-254-8888
>Ithaca, NY 14853-2602 Em: psz1 AT cornell DOT edu
>
>
>
>
--
Paul Zarnowski Ph: 607-255-4757
Paul Zarnowski Ph: 607-255-4757
747 Rhodes Hall, Cornell University Fx: 607-254-8888
Ithaca, NY 14853-2602 Em: psz1 AT cornell DOT edu
|
