Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Tailoring Cartridge Dispositions in Libraries to Customer Workloads

IP.com Disclosure Number: IPCOM000111421D
Original Publication Date: 1994-Feb-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 6 page(s) / 206K

Publishing Venue

IBM

Related People

Goldfeder, ME: AUTHOR [+4]

Abstract

Disclosed is a finite set of competing algorithms for deciding (a) when it is best to pre-emptively empty an inactive drive in an automated library of removable media in anticipation of the next mount request coming from an unmounted cartridge and (b) when it is best to leave inactive cartridges in their respective drives in anticipation of additional hits. Also examined is which inactive cartridge should be dismounted.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 37% of the total text.

Tailoring Cartridge Dispositions in Libraries to Customer Workloads

      Disclosed is a finite set of competing algorithms for deciding
(a) when it is best to pre-emptively empty an inactive drive in an
automated library of removable media in anticipation of the next
mount request coming from an unmounted cartridge and (b) when it is
best to leave inactive cartridges in their respective drives in
anticipation of additional hits.  Also examined is which inactive
cartridge should be dismounted.

      The competing algorithms are used for tracking a moving target,
the current workload of an automated library of removable media.
This workload could change gradually over weeks, as the library fills
up with data, or change hourly or faster between a random-retrieval
and a sequential mode.  These competing algorithms can be viewed as
logical strings (or vectors), as the values of elements at specific
positions in the string have specific meanings.  Sets of algorithms
are shown in Figs. 1 and 2.

      Although the library controller or host would use the
best-choice algorithm for making decisions, it would have an
internal-model of all competing algorithms.  This way, all of the
algorithms receive scores and are ranked in an order of merit based
on their success in predicting correct library actions.  Rankings
could be based on totally cumulative scores or a running average of
the last N I/O operations.  Preferred is a linear running average for
the ranking algorithm with the number of samples in the running
average equalling the number of drives in the library.  Although
noncompetitive algorithms could be discarded, there is no allowance
for such extinction because algorithms which are poor choices at one
moment maybe good choices later on.

      The logical strings shown in Figure 1 are expanded into
flowcharts in Figs. 3-5.  Fig. 1 is read as follows.  If there are
one or more drives empty in the library, String 0 is in effect and no
decision is made about KEEP, LRM, or LRU.  At any given time when
string 0 is not in effect, string 1, 2, 3, or 4 maybe the best
choice.  The four competing strings (1,2,3,4) are the following
cartridge dispositions:

1.  KEEP inactive cartridges mounted.  When a mount request comes,
    dismount the LRM inactive cartridge.

2.  KEEP inactive cartridges mounted.  When a mount request comes,
    dismount the LRU inactive cartridge.

3.  Preemptively unload the LRM inactive cartridge.

4.  Preemptively unload the LRU inactive cartridge.

      The goal is to quickly and continually explore to determine the
current best choice so that this best choice can be exploited and the
overall library response time be reduced.

      The following flowcharts show our implementation of Fig. 1.  If
there is a SIO (Start I/O or file-open) request in step 11 of Fig. 3,
the flowchart executes subroutine SIO, step 12.  If there is no
pending SIO request in step 11, the flowchart checks f...