Browse Prior Art Database

Hit Ratio Estimates for Non-Cached Volumes

IP.com Disclosure Number: IPCOM000112954D
Original Publication Date: 1994-Jun-01
Included in the Prior Art Database: 2005-Mar-27
Document File: 4 page(s) / 169K

Publishing Venue

IBM

Related People

Cord, JH: AUTHOR [+4]

Abstract

A method for determining when a volume in a cache DASD subsystem should resume caching once it has been put in a non-caching mode because its hit ratio is below a certain threshold is disclosed. The method to determine when to begin or recommence caching a volume uses the concept of track cache single reference residency time.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 28% of the total text.

Hit Ratio Estimates for Non-Cached Volumes

      A method for determining when a volume in a cache DASD
subsystem should resume caching once it has been put in a non-caching
mode because its hit ratio  is below a certain threshold is
disclosed.  The method to determine when to begin or recommence
caching a volume uses the concept of track cache single reference
residency time.

      The solution to the problem of knowing when to begin or
recommence caching a volume is to use the concept of track cache
single reference residency time, developed by [*].  Throughout this
document a volume is referred to as being cached or non-cached.  It
is important to realize that the term volume is used for convenience
only, and that the problem and solution apply equally to volumes and
to set(s) of contiguous DASD cylinders.

Using this concept, it is possible to,

1.  Decide whether to cache (or stage) a non-cached track when it is
    accessed by a channel program.

2.  Compute the estimated hit ratio for a non-cached volume, which we
    call the volume quasi hit ratio, and use it to decide whether to
    begin caching the non-cached volume.  The hit ration is the
    fraction of accesses to data that are satisfied completely from a

    RAM buffer (called a cache) and therefore did not require a delay
    to access the mechanical DASD in a cache DASD subsystem.

Track Cache Single Reference Residency Time Estimate

1.  Let q(i,t) be the t th instance of track cache single reference
    residency time  on volume i.

2.  q(i,t) is measured as the time in milliseconds (ms) for a cached
    track from the time a cache miss channel program attempts to
    access data in the cache subsystem  to the time the track is
    Least Recently Used (LRU) demoted from the cache.  Note that this
    measurement pertains only to track accesses that are misses, for
    which the staged track receives no hits.  Thus, it is a subset of
    all cache misses.

          The cache misses measured can be for read or write channel
    programs.  The classification can be further broken down into
    read miss residency time and write miss residency time, but this
    finer breakdown is not used in this document.

3.  A running sum, QSUM, is maintained of the series of track cache
    single reference residency times for the subsystem throughout the
    collection period.  The collection period is a fixed number of
    minutes, e.g., 15 minutes, at the end of which the average track
    cache single reference residency time is computed and the
    summation variables re-initialized.

    a.  NUMT=NUMT+1; where NUMT is the count of instances of track
        cache single reference residency times.

    b.  QSUM=QSUM+q(i,t);  running sum of the track cache single
        reference residency time instances.

4.  At the end of the summation period, the average track cache
  ...