We have a z/OS server task that takes requests from clients and does, like, 
things with them. The server manages a pool of TCBs. We have various controls 
on those TCBs, so a single rogue client can't just cause a zillion of 'em. But 
there are times that a client program gets a spike and hammers our server, 
which dutifully creates a bunch of TCBs.

The question arises when that spike abates: these TCBs currently remain idle 
more or less forever. We could add a mechanism to reap idle TCBs after some 
period, but the utility of this is unclear. Obviously if the spike does 
reoccur, they'll be useful.

So the question is: What's the real cost of having a TCB sitting there idle? My 
performance experience is with z/VM and Linux on z. Certainly on z/VM, an idle 
userid gets paged out and is essentially "free" (modulo a tiny bit of memory 
used for the VMDBK and page tables and like that). To what extent is an idle 
TCB on z/OS impacting things? Are there system- and/or primary TCB-based limits 
on numbers of TCBs that we're pushing against? Each TCB uses maybe 25K of 
memory in our address space, and presumably some small amount of system memory.

Thanks in advance to the collected wisdom of the group,

...phsiii

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to