Re: Question: Online 7.14 not shutting dwn completely

In article <r9fV1.100$yU.305044@newse1.midsouth.rr.com>, cbolden1@midsouth.rr.com (Carlos Bolden) wrote: > Hi All, > > I have a problem. We have a system at work ( NCR 3555 running > Informix > ODS 7.14, TOPEND 02.02.02.00 with a serious application built for it. > 186gig > of dataspace mirrored, and in need of some serious DB maintenance, 2 > gig of > ram and 16 pentium 133 processors, running NCR MP-RAS 3.00.02, Symbios > disk > arrays running Raid 5, ) > that has a nasty problem of: > > A) During shutdown, informix removes all shared memory segments, > seemingly > frees all memory ( We nail memory on this system for better performance > ) > but there are lagging oninit session hanging around on the system ( We > press this system really realy hard and it does a lot of IPC ) with it's > children in a zombie state. I thought that for MP-RAS the recommendation was for Non-forced residency. Also I believe you need very particular PBASE? releases. > B) In turn, Memory seems to slowly but surely creep away as the virtual > portion of informix memory slowly gets larger. They ( informix > consultants ) have said that there is a problem with the 'like' and 'as' > keywords and we have 5500 or so occurences of theses keywords built into > some of the primatives that we have defined. Sounds like you've got a leak. > > C) What's even weirder is that when he marks the instance as being > stopped, > Memory is not totally freed on the system. A large portion of swap ( > shared > memory ) and real memory ( backing store ) is slowly being freed it > eventuallt frees itself after about 45 minutes but it is very annoying > especially trying to schedule work to be done on this system during > system > downtime. How is this being detected? There is no reason for released memory to be identified immediately, unless you have another heavy process come along a try to use it, then as if by magic the memory should appear on the vmstat, swap etc. Remember the swap reporting tools all report swap on different collection criteria > > D) Understaning that NCR memory is not on a 4 meg boundary, the > phisyscal > ram chips are proprietary ( 246 meg on a chip or somthing like that ), > memory can potentially become very fragmented especially doing A LOT OF > INTERPROCESS COMUNICATION. AT ANY ONE TIME, WE MAY HAVE UPWARDS OF 500 > PROCESSES OR SO WITH A DATABASE CONNECTION AT LEAST 20 CURSORS OPEN AT > ANY > ONE TIME DOING WORK. ( SOME OF THESE PROCESSES GET PUT INTO THE > REALTIME > CLASS AND DO WORK IN MEMORY!!! ) WRT to the memory I spent most of last year working with NCR and one of the customers had a memory device failure. There was only one more available in the country, if we needed more then it had to come from the US but they were getting short of supply Paul Watson # I don't suffer from WF Software Ltd. # stress, I'm just Tel. (+44) 1436 674729 # a carrier Fax. (+44) 1436 678693 # www.wfsoftware.co.uk