Re: Informix 7.23.uc11 and HP10.2

Carlson@WHSmith wrote: > > 4GL: upgrade 6.02 to 7.20.UD1 > Engine: upgrade 7.23.uc4 to 7.23.uc11 > OS: upgrade HPUX 10.01 to HPUX 10.2 > > We've recently upgraded from HPUX 10.01 to 10.2 and are having some > major performance issues. I've noticed that the oninit processes will > not spin up to the challenge, so to speak. Most, if not all, processes > (4ge and shell) take longer and it appears that something is limiting > the oninits. > > For example, I run update stats after midnight on all databases. This > process normally takes 4 to 4 1/2 hours to run; it now takes 9 to 9 > 1/2. Normally my oninits will spin up to a higher percentage; this > morning, they may have bumped 30%. ( BTW, the update stats SQL is run > as a shell / SQL script from the command line. ) > > We have calls in to Informix and HP. Any possibilities from the > newsgroup? Here's what we've done in preparation for this upgrade: > > 1. We installed all recommended patches > 2. We've checked all kernel parameters acc. to release notes. > 3. We've verified that the SQL is taking the 'best' path to the > data. > 4. We've recompiled all 4ge programs with the 'new' compiler. > > Please feel free to crosspost this to any relevent HP newsgroups; I > wouldn't know which one to consider. > > Thanks in advance > Here's the followup: I had this weird idea days ago . . . . what about KAIO? I've heard that it either runs or it doesn't under HPUX. Under the original configuration (7.23.UC4 and HPUX 10.01) it gave us a bit of an increase in performance. I asked the HP SysAdmin about this, and he was assured by HP that nothing had changed, even though bug# 96414 seems to intimate that either HP's implementation changed or Informix's usage of KAIO changed. I was assured that nothing had changed. Until . . . . Late Friday night I get this strange idea. I disable KAIO and throw a bunch of AIO VPs out into the system, run some tests, and VOILA, our performance returns. I implement this on out production system and all seems to be well with the world. Only question . . . . Why would something beneficial like this work on earlier releases and not on a later release? John Carlson Informix DBA WH Smith, Inc. last seen: reviewing notes on AIO VPs . . . it's been so long. <G>