z/VM settings
This section describes the Quickdsp and SRM settings.
Quickdsp
The set Quickdsp command
and the quickdsp operand of the option directory
statement allow you to designate virtual machines that will not wait
in the eligible list when they have work to do.
All measurements
in this study that were run on z/VM® used
the quickdsp operand. For the guests it was specified
in the option directory statement for all z/VM virtual guest user directory definitions.
SRM settings
For some of
the tests, the z/VM SRM settings
were changed. The presence of the quickdsp option
directory statement, however, may make these changes less relevant
to the Linux™ virtual systems.
Other z/VM virtual users dispatching
priorities would be affected by the SRM settings because they did
not run with the quickdsp option in our test.
It is recommended to use CP SET SRM STORBUF to increase the z/VM system's tolerance for over-committing dynamic storage. CP SET SRM LDUBUF is often used to increase the z/VM system's tolerance for guests that induce paging.
Some of the tests used the SRM values shown in the following QUERY SRM command:
q srm
IABIAS : INTENSITY=90%; DURATION=2
LDUBUF : Q1=300% Q2=300% Q3=300%
STORBUF: Q1=300% Q2=300% Q3=300%
DSPBUF : Q1=32767 Q2=32767 Q3=32767
DISPATCHING MINOR TIMESLICE = 5 MS
MAXWSS : LIMIT=9999%
...... : PAGES=999999
XSTORE : 0%
LIMITHARD METHOD: DEADLINE
Ready; T=0.01/0.01 15:35:12LDUBUF parameters
specify the percentage of paging exposures the scheduler is to view
when considering adding or loading a user into the dispatch list with
a short, medium or long-running transaction, respectively. The values Q1, Q2,
and Q3 shown in the output above, refer to the expected
length of a transaction, where:Q3is the longest running transactionQ2includes medium and long length transaction usersQ1includes all users
The larger the percentage allocated, the more likely it is that a user is added to the dispatch list of users that are already on the list and waiting to be dispatched. Values over 100% indicate a tolerance for an overcommitment of paging DASD resources. A value of 300%, for example, indicates that all users in that transaction-length classification will be loaded into the dispatch list even if this would lead to an overuse of paging resources by up to three times.
The STORBUF parameters
are also specified in three values. The values specify the percentage
of pageable storage that can be overcommitted by the various classes
of users (Q1, Q2, Q3)
based on the length of their transactions, as described in the previous
paragraph. Again, Q1 includes all classes, and Q2 includes Q2 and Q3 users. Q3 is
reserved for long-running transaction users. Any value over 100% represents
a tolerance for that amount of storage overcommitment to users included
in that classification.
For some tests we changed the values using the SET command, as shown below, to set slightly less aggressive dispatching decisions for our non-Linux z/VM guests.
SET SRM LDUBUF 100 100 100
LDUBUF : Q1=100% Q2=100% Q3=100%
Ready; T=0.01/0.01 15:35:22
set srm storbuf 300 250 200
STORBUF: Q1=300% Q2=250% Q3=200%
Ready; T=0.01/0.01 15:36:19