2. PERFORMANCE REPORTING ANALYSIS › 2.4 I/O Configuration Analysis › 2.4.3 Shared DASD Analysis Inquiry › 2.4.3.2 Technique Tutorial

2.4.3.2 Technique Tutorial

DASD hardware allows multiple systems to be connected to a
single volume.  This feature is highly desirable if volumes
containing critical applications must be reached from another
system in the event of a processor failure, or to allow a
single copy of critical data to be shared by multiple
systems.

However, shared DASD often introduces significant performance
problems by allowing workloads on one system to adversely
affect activities on other systems.  Since RMF reports on the
I/O activities of individual systems rather than the I/O
activity of an installation as a whole, it is necessary to
summarize the RMF data from multiple systems to evaluate the
performance of a shared device.

Unfortunately, RMF does not contain data elements that
indicate whether or not a device is shared.  The algorithm
implemented by the Shared DASD Analysis inquiry assumes that
disk volume serial numbers are unique within the
installation.  Using this assumption, all of the records (for
the same time interval) for each volume serial number are
mapped together for the analysis.

The shared DASD analysis algorithm is based on data contained
in the CA MICS HARDVA File. The following values are computed
for each shared volume:

TOTAL SIOs:  The sum of the SIO count values, CA MICS data
             element DVASIO, for each of the SYSIDs in the
             configuration.

AV RESPONSE  The weighted average of the response times,
TIME:        data element DVAAVRTM, for all of the SYSIDs
             that share the device.  Response time is the sum
             of average device service time and the internal
             queuing for the device.  Internal queuing is
             defined as the time the I/O waited in the unit
             control block UCB and logical channel queues.
             This value is calculated by weighting the
             response time measured for each of the systems
             by the number of I/Os issued to the device by
             the system.

AV SERVICE   The weighted average of the service times,
TIME:        data element DVAAVSTM, for all of the SYSIDs
             that share the device.  The average service time
             is computed by dividing the device utilization
             by the start I/O rate.  The value is calculated
             by weighting the service time measured for each
             of the systems by the number of I/Os issued to
             the device by the system.

AV QUEUE     The weighted average of the internal queue times
TIME:        for the device (the difference of CA MICS data
             elements DVAAVRTM and DVAVSTM) for all of the
             SYSIDs that share the device.  The value is
             calculated using the weighted average service
             and response times.

% UTILI-     The sum of the percentages of the times that
ZATION:      each of the systems utilizes the device.  Prior
             to MVS/XA, utilization was defined as the
             maximum of either the percent of the time that a
             system used a device (i.e., device active,
             DVAPCACT) or the percent of the time that the
             system prevented other systems from using the
             device (i.e., device reserved, DVAPCRSV).  This
             definition was required, since RMF did not
             indicate the percent of time that a reserve
             condition existed when no I/Os were outstanding.

             Starting with MVS/XA, a new variable called
             DVAUTL was introduced that indicates the number
             of samples during which the device was reserved
             and no I/O was outstanding.  Hence, utilization
             in  an MVS/XA environment is defined as DVAPCACT
             plus the percent of time the DVAUTL condition
             existed.

The effect of the DASD sharing on each of the devices is
determined by calculating the following ratios:

    o  The ratio of each system's average response time to
       the weighted average response time for the
       configuration.

    o  The ratio of each system's average service time to the
       weighted average service time for the configuration.

    o  The ratio of each system's average internal queue time
       to the weighted average internal queue time for the
       configuration.

Each of these ratios is discussed in the following
paragraphs.

The ratio of the device response time on the system to the
average device response time for the configuration indicates
whether I/Os issued by the system are receiving better or
worse service than the average for the configuration.  Since
this value includes internal queuing times, it is an
indicator, but not the best index, for evaluating shared DASD
contention.

The ratio of the device service time on the system to the
average device service time for the configuration is the best
indicator of shared DASD contention.  It indicates whether
the system is impacting or being impacted by the I/Os issued
by the other systems.  Values less than or equal to one
indicate that a system is attempting to monopolize a device.
Values greater than one indicate that a SYSID is being
impacted by the other system.  For a well-balanced
configuration, all systems should either have values of zero
(indicating no utilization of the device) or values between
0.9 and 1.1.

The ratio of the device internal queue time on the system to
the average device internal queue time for the configuration
provides a relative indication of internal queuing on each of
the systems.  Values less than one indicate that the device
is lightly utilized on the system, while values greater than
one indicate that the internal queuing delay on the system
may be substantial.