2. PERFORMANCE REPORTING ANALYSIS › 2.3 Auxiliary Storage Management Analysis › 2.3.2 Usage Guidelines › 2.3.2.1 ASM Configuration Tuning
2.3.2.1 ASM Configuration Tuning
The interaction of the ASM configuration with the I/O
subsystem is not directly measured by the Auxiliary Storage
Management Analysis reports. The ASM Page/Swap Response Time
Report shows if service times are uniform or if the response
of specific data sets is causing a problem. The ASM Page/Swap
Activity Report can be used to evaluate the impact of not
isolating page or swap data sets on dedicated volumes. Both
data set response and service times are reported.
The following list gives some configuration guidelines that
are generally applicable to all levels of ASM support:
1. Three data sets is the minimum paging configuration.
The first data set defined is for PLPA, the second for
common, and the remaining data set(s) are for local data
sets. A minimum paging configuration of only three page
data sets (i.e., one PLPA, one common, one local) is not
very practical. Check page rate by device and data set
response times to see if more local page data sets are
required.
2. In pre-MVS/XA systems, use swap data sets in a TSO
environment or for an environment with heavy swapping.
Overflow of swapping into the local data sets adversely
affects the performance of the local data sets. In
MVS/XA or MVS/ESA systems, because of improvements in
the paging data set algorithms (as opposed to swap data
set algorithms), you should strongly consider the use
of local page data sets for swapping.
3. Define extra page and swap data sets (if used). Use the
PAGENUM keyword in the IEASYSnn member of SYS1.PARMLIB
to allow for the addition of extra data sets that can be
used to recover from an emergency condition that
exhausts auxiliary storage.
4. If the PLPA paging rate is very high, PLPA can be
allowed to spill into the common data set. The initial
configuration should be sized in accordance with
information presented in the IBM Initialization and
Tuning Guides. The WEEKS level of the ASM Page/Swap
Activity Report provides the information necessary to
make this decision.
If you use intentional PLPA overflow to common, be
careful to make the common data set large enough to hold
both the PLPA overflow and all common system area pages.
If the common data set is filled, the ASM stops using it
and the system enters a wait state.
5. Do not put ASM data sets on channels with non-RPS
(Rotational Position Sensing) activity. The non-RPS
activity will tend to dominate, because the ASM releases
the channel during sector searches. This condition may
be recognized by a build-up of the ASM queue.
6. If fixed head devices are available, allocate them to
the highest activity data sets. Most load modules are
normally placed in the link pack area because they are
frequently referenced, which means the PLPA data set is
a good candidate for a fixed head device.
7. Avoid placing more than one ASM data set on any single
moveable head device. ASM should be prevented from
competing with itself at the device level.
8. Avoid placing ASM data sets on any shared device. The
slot selection algorithms all depend on knowledge of the
arm position following the last ASM I/O request. ASM
channel programs cannot operate efficiently when another
CPU has moved the arm.
9. Over specify paging space. Larger common and local data
sets on moveable head devices increase the probability
that the slot selection algorithms will find slots
underneath or near the current arm position. Thus,
backing store requirements will not inhibit address
space creation or the allocation of virtual I/O (VIO)
data sets.
10. Use multiple local and swap data sets (if swap data sets
are used at all). Even when paging space or swapping
space requirements can be met by a single data set,
multiple data sets allow ASM to service multiple
requests concurrently. This is especially important
during peak loads.
11. Duplex Data Sets. The use of a duplex data set severely
affects common area page-out processing. You should
perform a periodic evaluation of the actual saving
(i.e., how many IPLs were saved) to determine if the
savings is worth the overhead.
12. Allocate entire volumes for ASM data sets, and place no
more than one ASM data set on a single path.
13. If real storage is constrained, reduce the amount of
logical swapping by lowering the maximum system think
time value on the LSCTMTE parameter in member IEAOPTEX
of SYS1.PARMLIB. See the Initialization and Tuning Guide
for a description of this parameter.