2. PERFORMANCE REPORTING ANALYSIS › 2.4 I/O Configuration Analysis › 2.4.7 Tape Drive Simulation Analysis Inquiry › 2.4.7.2 Technique Tutorial
2.4.7.2 Technique Tutorial
Tape drives, unlike DASD, are usually not shared between
systems. Generally, tape devices are physically attached
(cabled) to multiple systems, but are physically switched to
only one at a time.
Physical paths to tape devices are established by using
switches (toggles) on the control units. Both 3420 and 3480
control units have channel adapter switches. The 3420
control units also have separate toggle switches for each
device.
It is becoming more common to "share" the tape subsystem by
having some of the tape devices physically switched to two
systems but logically offline to one or the other. This
allows all systems access to the tape subsystem, while
preventing concurrent access to any single device. Software
access to the drives is accomplished by the use of the VARY
ONLINE operator command.
The 3480 tape subsystem can function in one of two modes:
o Full Function
o 3420 Compatibility
In Full Function mode, the drives are assigned a processor by
the use of the VARY ONLINE command. The command causes
values to be placed in the pathing array within the control
unit.
After assignment to a system, no other system can access the
drive unless the VARY command employs the SHR option. The
SHR option allows the drive to be assigned to multiple
systems. In this shared mode, tape devices must be
controlled by software (such as JES3 or other products) to
prevent concurrent access.
In 3420 Compatibility mode, pathing arrays do not exist. The
only way to prevent concurrent access is through the use of
the channel adapter switches.
As you can see from the above description, performance in the
area of batch throughput can be adversely affected not only
by the total number of tape drives available, but also by the
number of drives online to each system in a multiprocessor
configuration. Tape drive capacity and performance
prediction is an area often overlooked due to the complexity
and changeable nature of workloads and configurations in a
multiprocessor environment.
A common operational safeguard is to place the majority of
the tape devices on the primary tape processing system and
ensure that tape job class initiators are enabled on that
system. This prevents the physical switching of tape devices
from one system to another when a job or started task on
another system requests more tape devices than are available
to it. For this reason, you must take special care to ensure
that the true maximum tape drive limit is known for each
system prior to running the Tape Drive Simulation Analysis
inquiry. Generally, the inquiry should be run separately for
each processor in a multiprocessor complex.
As mentioned above, JES3 provides the capability to manage
tape devices which are online to more than one system, unlike
the standard MVS allocation scheme, which is only aware of
device status on one system. The JES3 facility that provides
device management is the Main Device Scheduler (MDS). MDS
attempts to satisfy the resource requirements of a job in the
context of a globally controlled environment, so that job
execution may proceed without allocation delays.
Standard MVS allocation is done at the step level as step
execution begins. MDS, on the other hand, considers the
requirements for each step of a job, within the context of
other currently scheduled work, prior to the start of job
execution. MDS reserves devices, thus making them
unavailable for other work, then allocates them at the step
level as in standard MVS.
RMF/SMF data sources do not currently report JES3
reservations, although the effect is similar to MVS
allocation in that the device is not available while it is
reserved. For this reason, MDS device requirements are
higher than MVS device requirements. On the positive side,
MDS minimizes tape device contention between jobs.
Under MVS allocation, job steps are competing for resources,
with the losers (those steps requesting unavailable
resources) either being cancelled by the operations staff or
waiting for the resource(s) to become available.
This waiting period is costly in that the waiting job is
holding other valuable resources such as an initiator, an
address space, data sets, and possibly devices. These held
resources can affect overall system utilization adversely.
MDS attempts to avoid this situation by not allowing jobs to
be passed to MVS for execution until their allocation
requirements can be met or "set up."
JES3 "setup" occurs while the job is in the JES3 address
space (the only resource used during setup is JES3 queuing
space). MDS reserves devices, then requests and verifies the
mounting of initial volumes on each device before the job can
be selected for execution. If "defer" is specified, volume
mount requests are not issued at setup.
The type of setup that MDS employs can be specified on the
//*MAIN JES3 control statement included in the job's JCL
stream, or via the JES3 STANDARDS initialization statement.
Using the initialization method is the common approach.
The types of setup that MDS can employ are:
o Job
o High Water Mark
o Explicit
Job or High Water Mark setup can be specified at JES3
initialization. Explicit is very similar to Job setup, but
can only be specified via the //*MAIN control statement.
These types of setup are discussed in detail below.
JOB SETUP
Job setup attempts to premount all volumes for all the steps
in a job prior to passing the job to MVS for initialization.
For example, a five-step job that requires three tape data
sets in each step causes MDS to reserve 15 tape devices,
ensure that these devices are mounted, and then pass the job
to MVS for initialization. Job setup is thus quite expensive
in terms of the drives required.
HIGH WATER MARK SETUP
High Water Mark setup is more efficient in device allocation.
Under this method, the greatest number of devices that any
step requires is determined prior to job execution, and that
number of devices is reserved for the job. In the above
example, three drives are reserved and the volumes required
by the first step premounted. Under both Job and High Water
Mark setup, drives no longer required are released as the
step ends.
A more complex example illustrates High Water Mark setup and
the releasing of drives. Job X has six steps, where step 1
requires one tape drive, step 2 requires five, step 3
requires no drives, step 4 requires four, step 5 requires
one, and step 6 requires two. This information is displayed
in the following chart:
STEP #
----------------------
1 2 3 4 5 6
----------------------
Drive Requirements 1 5 0 4 1 2
High Water Mark Reservations 5 5 4 4 2 2
Job Setup Reservations 13 12 7 7 3 2
----------------------
As shown above, High Water Mark setup ensures that steps will
not wait for device allocation after execution begins, and is
much more efficient in its use of tape devices. The Drive
Requirements line of the chart can be equated to standard MVS
allocation if the drives are available when the step begins
execution.
EXPLICIT SETUP
A third method, known as Explicit setup, can be specified via
the //*MAIN JES3 control statement. Only the premounting of
drives is affected under Explicit setup. Device reservation
defaults to the method used in Job setup.
The type of setup to employ or whether to employ MDS setup is
dependent on your installation's needs and requirements. For
more information, consult the following IBM manuals: JES3
System Programming Library: Installation, Planning, and
Tuning; JES3 Introduction; and JES Overview.
The Tape Drive Simulation Analysis inquiry allows you to
examine the utilization of your tape drive subsystems during
periods that have significant tape drive availability
problems. It also permits you to develop simple models to
analyze the effects on the tape subsystem of moving or adding
workloads. By calculating the allocation of tape drives
repeatedly over sufficiently short intervals, this inquiry
can aid you in ensuring efficient tape drive utilization when
scheduling existing or new work.
You can use Tape Drive Analysis inquiry for the following
purposes:
o Provide graphical and tabular reports on the status of
the tape subsystem.
o Model the effects of change on the subsystem.
o Aid in capacity planning and performance tuning of the
tape subsystem.
The Tape Drive Simulation Analysis inquiry provides two
algorithms for reporting tape drive allocation levels. One
provides a precise snapshot of drive allocation levels at
each measured time interval. The other prorates drive usage
according to the percentage of the interval during which any
particular job step was executing.
The first method may be most helpful when you want to obtain
a moment-by-moment report of tape drive allocation patterns.
The second method is particularly helpful in answering
capacity-related questions.
The following sections discuss the operation of the Tape
Drive Analysis inquiry in terms of:
1 - Data Limitations
2 - Computational Methods Available
3 - Analyzing JES2 and JES3 Drive Allocation
4 - Exception Reporting
5 - Modeling
6 - User Modeling Exit
