Eric Thomas <ERIC@FRECP11>
Sat, 30 Jan 88 18:07:00 SET
|
I'm a bit worried about this prime time aspect, I mean, how can it be
implemented? Given that:
- A global signoff operation is going to take a lot of CPU time anyway. You're
going to be presented with a few hundred accounts to be removed from your
10-20 lists, and it's probably going to happen at the same time, give or
take a few weeks, for all the US universities.
- There is no way to hold a global signoff DISTRIBUTE job during prime time
without improving DISTRIBUTE, that is DIST2, and that you have a problem if
one of the server in the path is not up to date, etc. I mean, I can hold the
job during prime time at the source node. But once it's sent, it will be
DISTRIBUTEd immediately by the other servers (again, unless a PRIME= keyword
is added to DIST2).
- There is no problem about providing a PRIME= keyword on the SIGNOFF command,
and maybe even forcing it on if the request is a SIGNOFF * and blah and etc.
But then you're saying that you're only worried about the CPU time the
SIGNOFF itself will take, not the CPU time required for the DISTRIBUTE
operation.
We only have a choice between putting the restriction in DIST2 and putting it
in SIGNOFF, which has to be updated anyway to provide a FOR option (you can of
course issue 200 FOR userid SIGNOFF *, but a single SIGNOFF * FOR DD=NAMES
would allow me to use a more efficient algorithm which would save a lot of
CPU, and in this case the QUIET option could be honoured).
Eric
|
|
|