🚀 go-pugleaf

I am looking for some information/perspective on
system accounting questions.

I work with a Cray Y-MP, Cray II, and soon to be with a C90,
and there is a minor accounting question that has been raised.

Every year there is the annual allocation of 'cpu time' to
the various groups. Including the ubiquitous 'over_head'
for 'what it takes the system to do its job.'

What I am looking for are means to get a 'grip' on what
these 'overhead' factors really are.

The basic idea is that there are 'N' units of time
when the system is up in multiuser mode, and we need
to account where those seconds go. Anytime a job 
finishes, a 'pacct' record is written out. So those
I have no problem with. The question gets a bit sticky
with those processes, like /vmunix that come 'up' prior
to accounting starting, and are still running AFTER 
accounting is shut off.

I have joked that a solution would be to run /vmunix
under time: 

	BigMachine: 1> time /vmunix

So that we can get a sense of how much 'usr/sys' time
the kernel actually takes to do its job....

If anyone has any ideas/suggestion/comments on better/useful
ways of 'bean counting' the actual 'system overhead' that
goes into supporting End_Users, I will appreciate any
comments or kibbitzing.

ciao
drieux

ps: I'll even take creative flaming if it helps me
get a better idea!





-- 
Modern Day Problems:
Sorry Sweatheart, But Just Because you were a BubbleHead, doesn't Count;
'boats' are not ships, and do not count for 'Ship Sinking.'
If You have to ask, look for an Adult, PLEASE.