Re: ODS 7.23UC5 and HP-UX 10.10
Posted in 1997
In article <3455fd6d.323789390@news.usr.com>, Jeff Kalchik
<jkalchik@spam-me-not.usr.com> writes
>Hi all,
>
>I'm in the process of bringing up an Informix engine (7.23UC5) on an
>HP9000 K420 (HP-UX 10.10, 2 processors, 1 gb RAM, Model 20 RAID
>ARRAY). Aside from an shmat error (which I may have solved with
>Informix Tech Support, increase SHMMAX), this has been going rather
>well. I've gotten the distinct impression from lurking here that
>there are some definite performance problems with HP-UX 10.10. Can
>someone please expand on this, and is it a good idea to upgrade to
>10.20? I've got the media, but have not installed it yet.
>
>A quick tuning question as well: assuming 2 processors,
>
>MULTIPROCESSOR 1 # 0 for single-processor, 1 for>multi-processor
>NUMCPUVPS 1 # Number of user (cpu) vps
>SINGLE_CPU_VP 1 # If non-zero, limit number of cpu vps>to one
>NUMAIOVPS 2 # Number of IO vps>
>I'm using these for testing purposes, and will be using KAIO in
>production. I'd assume that the only change I should make from these
>would be to set SINGLE_CPU_VP to 0.
>
I would set
MUTLIPROCESSOR 1 always as you have >1 cpu.
NUMAIOVPS 1 if you are using KAIO AND all your chunks are on raw
character devices.
The only question is :-
NUMCPUVPS 1 # Number of user (cpu) vps
SINGLE_CPU_VP 1 # If non-zero, limit number of cpu vpsto one
or
NUMCPUVPS 2 # Number of user (cpu) vps
SINGLE_CPU_VP 0 # If non-zero, limit number of cpu vpsto one
and the answer is try it both ways and see which works the best. It is
whether the overhead of the two CPU VPs syncronising is less than the
extra benefit you get with another CPU VP. i.e. being able to use both
CPUs at the same time.
This will depend upon the hardware,efficiency of the system calls
online uses on your version of UNIX and exactly what is being run. Too
many variables to get a definiate answer. Try it both ways 1 CPU VP or
2.
>TIA for all assistance.
>
>Jeff Kalchik
>
>----------------------------------------
>Views expressed are my own, not necessarily
>3COM's.
--
David Williams