ADSM-L

Re: Questions on the mechanics of 3590 tape data access

1996-07-10 12:47:16
Subject: Re: Questions on the mechanics of 3590 tape data access
From: Chris Hopkins <cjhopkin AT ON.BELL DOT CA>
Date: Wed, 10 Jul 1996 12:47:16 -0400
When I first started using the 3590 with ADSM I had much of the same
concerns as you have highlighted here.  I am running ADSM on a RS6000 AIX
box and am using the API to communicate with ADSM from within my
applications.  At one point I was sufficiently unhappy with ADSM's
performance (I have made changes since that have improved things
dramatically) that I started writing my own manager directly to the 3590
tape driver.  This worked quite well, but I can't replace the 1000s of
person months put into ADSM within a few weeks and as such I scrapped the
idea once I found better ways to use the ADSM API.

I did find that the 3590 is a great performer as far as tape devices go.
When talking directly to tape, I was able to randomly retrieve 100KB files
off a 10GB tape with reasonable performance.  If I was careful how I sorted
files before reading them, I could achieve consistently good performance.
The trick was I needed to be aware of the way in which the 3590 writes to
tape.  The following docs helped significantly:

  IBM SCSI Tape Drive , Medium Changer and Library Device Drivers -
Installation and User's guide. (GC35-0154-01)

  Magstar and IBM 3590 High performance Tape SubSystem Technical Guide (Red
Book GC24-2506-00)

From my experimenting, I am fairly sure that ADSM does not make the best use
of the 3590 capabilities, nor the dual gripper feature of the 3494
Library...  I had heard (some time ago) that IBM was focusing on performance
issues including the 3590 in their next release... Perhaps some release
insight would help here...


At 10:28 AM 7/10/96 -0400, you wrote:
>    I'm seeking insights into just how data is accessed within a 3590 tape,
>and more specifically, how ADSM accesses it.
>    We currently have an AIX system with a client and server on that same
>host, with fast/wide SCSI connections to 3590 tape drives in a 3494 tape
>robot.  All data (some 10 GB) is currently contained on one 3590 tape.
>We find that in retrieving an HSM-managed file (small) which has been migrated
>to tape that it takes a nominal 3 minutes to get the file back to disk, that
>time including the 32 seconds that it takes for the tape mount to complete.
>    More recently we had an experience where we needed to retrieve a bunch of
>files, averaging 1MB in size, to copy to a non-ADSM 8mm tape.  All of this
>data is on a single 3590 tape, as stored by ADSM, and in total is some 1.5GB.
>We started that job yesterday and, some 22 hours later, it is less than half
>done.  The 3590 tape has remained mounted all the time.
>    It's obvious that tape, be it 3590 or any other kind, makes a poor random
>access medium.  We just didn't expect it to be this bad - almost
>debilitatingly bad.  These experiences beg some questions which someone may
>have answers to...
>
>1. Just how does data access occur on a 3590 tape?
>   The usual 3590 manuals go no further than quoting specs (128 data tracks on
>   a tape, accessed 16 at a time, making for 8 pathways; plus servo tracks;
>   etc.).  Does anyone know of a written description, preferably available
>   online, which delves into the mechanics of going after data on a 3590?
>   (Can a host program, for example, take a shortcut to one of the pathways
>   rather than plow through the whole tape to get to a given file.  Is there
>   any kind of direct addressability on 3590s?  Etc.)
>2. Just how does ADSM manage data on 3590 tapes?
>   I can find no information on this.  Does ADSM store database information as
>   to where on a tape a given file lives, or is it limited to knowing that
>   it's somewhere on a given tape?
>3. Is there a way to get from ADSM a list of the files as they are stored on
>   the tape?
>   If we could get this we could go after the files in the order in which they
>   appear on the tape and make retrieval sequential instead of random, thus
>   dramatically reducing retrieval time (and drive monopolization and tape
>   head wear, etc.).
>
>    Any information in these areas would be appreciated.
>             Richard Sims, Boston University OIT
>
>
----------------------------------------------------------
Chris Hopkins                          cjhopkin AT on.bell DOT ca
Chris Hopkins                          cjhopkin AT on.bell DOT ca
Sr. System Designer
Highway 407 Electronic Toll Collection System
Bell Sygma Telecom Solutions
Toronto: (416) 215-2910
Ottawa:  (613) 781-4544
----------------------------------------------------------
=========================================================================
<Prev in Thread] Current Thread [Next in Thread>