Re: IDS 9.30 on a virtualised HP-UX platform; "Checkpoint Stalled"

Well...I tried going to single CPU VP, and it stalled almost immediately so I thought "Oh what the hell!" and set it to 8 CPU VPs (there are 8 on the virtualised machine), along with150 AIO VPs. It then went a lot further than before but then just stopped. Not at a checkpoint or anything, just all activity was dead (as shown in onstat -u, onstat -g glo, onstat -l, onstat -g ath, anything whatsoever). onstat -g act just showed a polling thread. I then finally managed to get the client to set up KAIO (there was a problem in that you can't grant MLOCK to informix in a virtualised container on HP-UX, it has to be done globally on the host. Outside my pay grade to understand this...) and when we finally got that running lo and behold, it loads up fine (and VERY quick!) So, I'd imagine there's an issue - but remember this is a VERY old IDS but $PARENT_COMPANY is not going to be spending any money upgrading, getting a renewed support contract and so on - with AIO on virtualisation, somehow, somewhere... I'm sure somewhere else along the line when we try to clobber it with 100+ users doing DSS and OLTP concurrently, something else will show up too. But, as they say, "may you live in interesting times!" Thanks Fernando, Art, for looking at this. On 19/02/2013 00:43, Fernando Nunes wrote: > I'm taking from what I remember... The situation you describe matches > what I remember. It was a problem on CREATE INDEX. That's probably why > you see the "X" in the flags... It's in a critical section. It should > be in/out instantaneously but due to a timing issue it never gets out. > The checkpoint cannot complete while there is any thread in a critical > section. That's why you see the checkpoint hang... > > This, and the other issue I had, can be considered a bug in > Informix... question is, nearly no customer ever hit them on the "real > thing", and it seems that on the virtual environment it's 100% of the > time. I think one of them was solved in 9.x something... So it could > help if you move to the latest fixpack/patch ever built for that > version on PA-RISC. But I'd have to dig up some infomation to check > this. As I mentioned I didn't follow up the situation to the end... I > was overloaded at the time and a Spanish colleague took the situation. > I can try to check. > But in short, given these situations and the one I heard from Oracle > it seems that the emulation is not like the real thing. I suppose you > considered moving to a newer version on IA64? It's really a waste of > technology to use IA64 to emulate old stuff (and trust me, I'm not a > very big fan of IA64). > If you have maintenance/support active, I think you could try to open > a PMR. Explain that you're trying to move... I believe Support Pilot > would apply to that situation and you could get the latest > fixpack/patch for the old platform. It's possible that would help, but > tech support should be able to find the bugs. That first one is pretty > easy to match. The other one was not that easy I believe. > > Regards. >