MANAGER@SMITH.BITNET (Mary Malmros) (07/21/87)
I'm thinking about increasing TQElms on my system, and need some information. 1. I am assuming that the way to determine if TQElm should be increased is if a process doing non-compute-intensive stuff keeps running out of quantum and getting requeued. Is this assumption correct? 2. Does anyone know of a system service that can be used to tell if a TQE has happened? Thanks in advance. +++++++++++++++++++++++++++++++++++ MMMMM MMMMM 2222 + + MM MM MM MM 222 222 + Mary Malmros + MM MM MM MM 22 222 + Systems Manager + MM MM MM MM 222 + Center for Academic Computing + MM MM MM MM 2222 + Smith College + MM MM MM MM 22222222222 + Northampton, MA 01063 + MM MMMM MM + (413) 584-2700 x3073 + MM MM MM + MANAGER@SMITH.BITNET + MM MM + + MM MM +++++++++++++++++++++++++++++++++++ MMMM MMMM
LEICHTER-JERRY@YALE.ARPA (07/23/87)
I'm thinking about increasing TQElms on my system, and need some informa-
tion.
1. I am assuming that the way to determine if TQElm should be increased
is if a process doing non-compute-intensive stuff keeps running out of
quantum and getting requeued. Is this assumption correct?
2. Does anyone know of a system service that can be used to tell if a
TQE has happened?
TQElm is a limit on the number of TQE's - Timer Queue Elements - a process
may have allocated. TQE's are "things", not events; it's meaningless to
talk about a TQE as "having happened". TQE's are used to store requests for
timer-based services - for example, SYS$SETIMER allows you to request that
an AST be delivered to your process at a given time. TQElm is also debited
for a couple of other random things, none of which I can remember at the
moment.
TQE's have nothing whatsoever to do with quantum expiration. You cannot
directly influence quantum-end events on a process-by-process or username-
by-username basis. The only thing you can change is the size of the quantum
itself. This change affects the entire system. In broad terms, larger values
for this parameter increase overall system throughput at the expense of
interactive responsiveness, while smaller values increase responsiveness at
the expense of throughput. "The default value is usually adequate." Small
changes will be unlikely to have any noticable effect on your system; large
changes are much more likely to screw your system up than to help, unless you
are running a very unusual job mix.
I STRONGLY suggest that you read and understand the "Guide To Performance
Management" before you start trying to "tune" your system at this level.
-- Jerry
-------