Phil Smith III wrote:
Les Koehler wrote:
The gory details escape me now, but we had a customer application at IGS that
issued a *lot* of CP commands. Customers complained of poor response time so we
consulted with Development and were advised that in that particular situation
we should abbreviate CP commands for improved performance.
We were doubtful but coded an old vs. new test and timed their performance
(both Virtual and Elapsed) on our lightly loaded Customer Production cpu at 2am
on a Sunday when we had a maintenance window and customers couldn't log on.
The results confirmed what Development had said, so we retested on our highly
loaded Dev system during prime time and the resulting difference was obvious.
That was sometime long before my retirement in May 2004 and was a *very*
unusual case. Generally, I would always use the fully spelled out CP command.
Les,
I know you well enough to trust your sincerity, but empiricism beats out memory
any day. I just ran a test and the full command was consistently faster, trying
various CP commands. So...color me skeptical. I'm trying to remember if there
are CP commands besides IPL whose behavior varies depending on whether they're
abbreviated or not; I can't think of any. Anyone?
...phsiii
My admittedly faulty memory tells me it had something to do
with querying spool files. Our offering, SHOWBBS, now
withdrawn, used 'pooled' userids and 'spool servers'. One
server got all the spool files for all the poolids and
dealt them out to the other spool servers after hiding the
pseudo-id in one of the extended attributes. This avoided
the spool file per user limit that CP had.
When the user logged on to SHOWBBS and was given one of the
poolids, the central spool server was notified and it
passed the pseudo-id and poolid to the other spool servers.
They then queried their spool to find all the spool files
belonging to the pseudo-id, retagged them for the poolid
and transferred them to the user.
When the user went to view his mail, our code had to do
some trickery to make everything look 'normal' in the
dialog the user saw.
At logoff, all his spool files had to be redistributed to
the spool servers.
As you can see, all this required a lot of CP commands and
*somewhere* in there we made the change that helped. It was
a long time ago, so I won't recommend that trick anymore!
Apologies for all the technical details. SHOWBBS was well
accepted by customers and an interesting project to lead.
The folks that originally developed it wrote some great
Rexx code and the two contractors that took it over and I
worked with, Will Rodgers and John Shaw (now deceased) were
certainly Best of Breed!
Les
