Xref: utzoo comp.sys.hp:4945 comp.unix.wizards:21762
Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!samsung!munnari.oz.au!kaukau.comp.vuw.ac.nz!comp.vuw.ac.nz!dsiramd!actrix!paul
From: paul@actrix.co.nz (Paul Gillingwater)
Newsgroups: comp.sys.hp,comp.unix.wizards
Subject: Is this a record?  kilo-semops/s!
Message-ID: <1990May2.092605.16728@actrix.co.nz>
Date: 2 May 90 09:26:05 GMT
Reply-To: paul@actrix.co.nz (Paul Gillingwater)
Organization: Actrix Public Access UNIX, Wellington, New Zealand
Lines: 24

I wonder if this is a record....  I've been trying to track down
a problem (or three) on a system running two major software applications
for over 200 users.  The system is an HP9000/855, running HP-UX 3.1,
and the applications are based on Informix C-ISAM and MSM.  The record?

Would you believe sar -m reports over 7,000 semops per second!
And that's a sustained average for some hours..  :-(

Hey INFORMIX!! (Sorry for shouting).  Any chance of supplying
a port of C-ISAM that uses a different semaphore implementation?

Net gurus:  does anyone have some ideas about what happens when you
have hundreds of users (yes, really) fighting for hundreds of 
semaphore structures at the same time?  (shared memory).
How can we predict the response degradation?  Someone (Joe Talmadge)
suggested there's an n-squared time to wakeup for processes waiting
on semaphores.  How can we correct this?  Do we just throw more
semaphores into the kernel?

Any input would be welcomed.


-- 
Paul Gillingwater, paul@actrix.co.nz