2. PERFORMANCE REPORTING ANALYSIS › 2.4 I/O Configuration Analysis › 2.4.4 DASD I/O Component Analysis Inquiry › 2.4.4.2 Technique Tutorial

2.4.4.2 Technique Tutorial

The I/O response time is the most significant part of
response or turnaround time for all but the most CPU bound
applications.  Over the past decade, we have witnessed CPU
internal processing speeds increase 10 to 20 times.  However,
the average response time of a 3390 is only about two and
one-half times faster than the 2314, which was initially
introduced with the IBM S/360 series of processors.  These
trends in device and processor speeds only serve to
exacerbate the performance problems that result from I/O
delays.

MVS/XA AND MVS/ESA

Under MVS/XA and MVS/ESA, the components of I/O time are
physically measured by the hardware.  I/O response time is
divided into the following components:

  o  Queue time - The  time  that the I/O was queued  waiting
     for the path or the device.

  o  Pending  time  -  The  time  after  the I/O  was issued,
     before it initially connected to the device.

  o  Disconnect time - The time that the device was busy with
     seek, latency, and missed RPS rotations.

  o  Connect time - The time that the device was connected
     to the path in data transfer mode.

  o  ESCON port busy time - The time that the device was
     delayed due to an ES Connection Director (ESCD) port
     being busy.

MVS/370

In environments prior to MVS/XA, the I/O response time as
perceived by the user application or transactions can be
calculated as the sum of a number of components.  By breaking
response time into components, you can identify problems that
can be addressed individually in order to improve
performance.  Two analytic methods have been published for
dividing I/O response into component parts:

  o  A technique that was published by Dorn, Friesenborg, and
     and Hallsworth that breaks I/O response time into four
     components.

  o  A technique presented by Sprinzen that breaks I/O
     response time into eight components.

Although the second technique is more detailed, it requires
that a number of assumptions be satisfied and that detailed
knowledge be available about the switching and composition of
I/O strings.  RMF data do not satisfy the necessary
requirements for this technique.  Hence, I/O component
analysis uses the first method.  Neither of these two
analytic methods is exact; however, they both provide
valuable insight into the delays encountered by DASD SIOs.

The interpretation of these values is discussed in the
following sections:

    1 - I/O Component Analysis for MVS/XA and MVS/ESA
    2 - I/O Component Analysis for MVS/370