2. PERFORMANCE REPORTING ANALYSIS › 2.3 Auxiliary Storage Management Analysis › 2.3.4 MVS Concepts › 2.3.4.2 Paging
2.3.4.2 Paging
ASM defines three classes of paging space data sets,
corresponding to the three logical groups of pages (pageable
link pack area (PLPA), common, and local).
At least three paging data sets must be defined to the
system. The first one is the PLPA data set, the second is
the common data set, and the remaining data set(s) are used
as local page data sets. You can define a fourth class of
paging data set, the duplex data set, to provide a backup for
the PLPA and common data set. Section 2.3.4 outlines what
pages will be written to each of these data sets. In this
section we discuss some device and data set size
considerations.
The term slot is used to characterize the location of a
virtual page when it resides on an ASM data set, just as the
term frame is used when a virtual page resides in real
memory. When a page is resident on an ASM data set, it is
said to occupy a slot. ASM paging data sets are sized in
units of slots.
PLPA DATA SET
The PLPA data set should be on a fixed-head device if
possible. If not, make sure this data set is on a dedicated
volume. Generally, a good starting point for sizing the PLPA
data set is four megabytes. In MVS/XA systems, because of
the potentially very large extended PLPA, this eventually may
not be large enough, but it is a good place to start. If the
PLPA data set is too small, it will spill into the common
data set. Although this is not recommended, we give an
example of how this might happen in Section 2.3.2.1.
COMMON DATA SET
The common data set should be large enough to hold all common
system area and nucleus area pages (in MVS/XA the nucleus is
conceptually included in the common area). As we noted for
the PLPA data set, a reasonable starting point might be four
megabytes (the same MVS/XA considerations apply), then adjust
the size to the actual requirements.
When the common data set is too small, it will overflow to
the PLPA data set. If this happens, ASM will not use the
common data set again, even when slots become available. For
this reason, you should avoid overflow from the common paging
data set to the PLPA data set.
DUPLEX DATA SET
The duplex data set should be large enough to hold both PLPA
and common pages. Thus, it should be as large as the
combined size of the PLPA and common page data sets.
Generally, eight megabytes should be sufficient for systems
through MVS/ESA. Again, you can use the Auxiliary Storage
Management reports to determine size requirements for this
data set.
Use of duplexing is not generally recommended because of the
extra overhead involved in the duplex operation (an extra
page-out is done for each common area page-out). If you use
duplexing, we recommend that you evaluate the savings in
terms of system down time to see if this option is still
required.
LOCAL DATA SET
The total paging space provided by the local data sets must
be sufficient to hold all private area pages and virtual I/O
(VIO) data sets. An example of this size calculation is
given in the Initialization and Tuning Guides. You can use
the Auxiliary Storage Management reports at the monthly level
to refine this value.
The local data sets should be sized to reflect the type and
number of address spaces expected to be active concurrently,
the private area for each, and the total amount of VIO pages
required. System and started task address spaces require
local page data set slots equivalent to their private area
size. To help ensure that there will be enough ASM slots
(sometimes referred to as backing slots or backing store)
available for the address space paging, swapping, and VIO
demands, MVS performs the function described below.
In pre-MVS/XA systems, at address space creation for TSO and
batch, ASM divides the number of slots that would be required
to contain the entire private area of the address space by a
constant (ILRSLOTC, currently initialized to eight) and
subtracts the result from the slot count for all local page
data sets. The number of slots required for a VIO data set
is divided by another constant (ILRSLOTV, currently
initialized to four) and, again, the result is subtracted
from the overall slot count. When there are no more slots
available by this computation, further address spaces or VIO
data sets are not allowed.
In MVS/XA and MVS/ESA systems, these functions are
implemented through the ASMSLOTC (for non-VIO requirements)
and ASMSLOTV (for VIO requirements) values. The calculations
are as described for pre-MVS/XA systems, but the default
assumptions are different. It is assumed in both MVS/XA and
MVS/ESA systems that you have allocated enough paging space
to accommodate the needs of your workload. Hence the default
values for both parameters effectively turn off the backing
store functions. The method of turning them on and adjusting
them to the needs of your workload are described in the
Initialization and Tuning Guide.
This algorithm is NOT based on the actual number of local
paging data set slots currently in use. Instead, it attempts
to predict the number of slots that will be used by an
address space or VIO data set and effectively reserves that
number of slots.
The PLPA and common data sets should be sized somewhat above
the minimum requirements that can be calculated. A growth
factor can be determined from the ASM Page/Swap Activity
Report and should be included. This growth factor is
typically 25 percent to 35 percent of the maximum slots used
over the last four months.
Sizing of local data sets is somewhat more complex. If the
data set is going to service group requests (swapping, VIO,
virtual fetch), it should be large enough to make efficient
use of the consecutive slot allocation.
In sizing the PLPA and common data sets, you should be aware
that overflow from one data set to the other is allowed. In
general, only overflow from PLPA to common should be
accepted.
The following sections discuss details of the paging
algorithms used. Section 2.3.2 explains the application of
these ideas.
1 - Data Set Selection
2 - Service Burst Calculation
