ADSM-L

Re: Drivers for 3590's

1996-12-05 13:40:03
Subject: Re: Drivers for 3590's
From: Michael Fink <Michael.Fink AT UIBK.AC DOT AT>
Date: Thu, 5 Dec 1996 19:40:03 +0100
On Thu, 5 Dec 1996, BELL, CRAIG wrote:

> There are a few applications, mostly using the API client,  that are heavy on
> the read/restore side of tape usage.  If the application is reading
> a number of
> files from the tape, AND the files are small relative to the tape's capacity
> (eg, 25MB), AND they are randomly spread over the tape, then each file must
> still be located on the tape in the same order that they were written to the
> tape [...] In a worse case
> scenario, you could rewind several times.

> There isn't an easy fix for this, but neither does it describe most customers'
> applications and needs, especially for normal backup and archive work.

Do you mean "most ADSM customers" or "most 3590 customers"? I doubt that
retrievals are always sequential for typical ADSM loads, in particular
when you look at HSM, which may have very demanding access patterns
(simultaneous requests of many files by many users of a system).

Provided that a device driver does support rapid positioning, scheduling
of random tape access is not a device driver issue, but can only be
addressed at the application program (i.e. ADSM) level.

> Perhaps those with experience doing that could post some of their
> results, along with some of the throughput rates they are getting on 3590s.

I want to point out that the question of performance optimization
of serpentine tape drives has been addressed in a very interesting paper
by Hillyer and Silberschatz at Bell-Labs (19 references). The authors
have developed a model for locate time behaviour and a scheduler for
DLT drives and report an improvement from 50 (unoptimized) to nearly 400
random retrievals in a DLT4000. The results should be easily generalized
to other types of tape drives.

I think that ADSM development should not neglect this important piece
of research.

Since I have lost the original anonymous ftp location of this paper,
I have made it available to the ADSM community in a hidden directory
in our university's anonymous ftp server:

  ftp://ftp.uibk.ac.at/pv/silberschatz/serpentine.opti.locate.ps
  ftp://ftp.uibk.ac.at/pv/silberschatz/serpentine.opti.locate-appx.txt

I'd be grateful if anyone still has the pointer to the original location.

Sincerely,

   Dr. Michael Fink +-----------------------------+------------------------
        EDV-Zentrum | Universitaet Innsbruck      | Tel: +43(512)507-2311
 Computing Services | Technikerstrasse 13         | FAX: +43(512)507-2944
--------------------+ A - 6020 Innsbruck, Austria | Michael.Fink AT uibk.ac DOT 
at
=========================================================================
<Prev in Thread] Current Thread [Next in Thread>