Re: NOTRANS processes and hosed up disks.

Ed Hobbs (ewh@alert.com) wrote: : Background: We're running on Online 5.00, IStar 5.00 and 4gl/sql 4.10 : on RS/6000s. Our rootdbs is 600 MBytes striped accros 6 different : disks. We also have a separate dbspace for tables which is some 9 : GBytes in size. The logical logs are in a separate dbspace but the : physical log is in rootdbs. : Problem: During periods of heavy activity, processes in a NOTRANS : state will hang-around for minutes before exiting. When this happens : the number of NOTRANS processes creeps up to a some arbitrary value : then severly drops to a low value almost instantly. The higher the : value gets, the slower the system gets. Its like they are being : blocked by something. During the last incident I also noticed that a : particular disk was being hit at 100% capacity (iostat out: %tm_acct : was at 100%), and that this disk contained a portion of rootdbs. Ed, Run this script and see exactly where your disk activity is occurring: # #watch_hot while ( 1 ) $INFORMIXDIR/bin/tbstat -z sleep 5 $INFORMIXDIR/bin/tbstat -D | grep / end NOTE: This script will zero out your profile statistic every 5 seconds, so if you are keeping any stats, this will screw them up!!! Look at where your disk activity is occurring. I'll bet that you are pushing your physical log in rootdbs pretty hard. If you are getting absurdly high writes in rootdbs, you may have found the problem. Also look at (maybe before running watch_hot) the lokwaits and lchwaits parameters in tbstat -p. If these are high, you may be running into contention for locks and latches. Of course, Dave's suggestion about checkpoints is absolutely correct. Look at the checkpoint interval in your online.log and see how fast they're occurring. : Question: I dont know how Informix manages it raw dbspaces. If a : portion of rootdbs is on a hosed up disk, can this cause a bottle_neck : for access to later sections of that same rootdbs. And if so, could : processes not be exiting because they cannot release locks to temp : tables contained on these unreachable sections of rootdbs. I doubt that portions of the disk are physically hosed. 5.0 is picky about this and would probably throw the chunks off line. : -- : / __ \\/ /__ ___ / /_____ ____ ____ ___ Ed Hobbs : / /_/ / / _ \\/ __/ __/ ___/ __ \\/ __ `__ \\ Database Administrator : / __ / / __/ / / /_/ /__/ /_/ / / / / / / Alert Centre, Inc. : /_/ /_/_/\\___/_/ \\_(_)___/\\____/_/ /_/ /_/ ewh@alert.com/(303)488-7636 -- =========================================================================== jlumbley@netcom.com (Joe Lumbley) "If you believe that my statements Database Administrator represent my company, please send The Tigon Corporation money and I'll send you stock" Dallas, Texas Watch for my _INFORMIX DBA SURVIVAL GUIDE_ in Fall '94 from Prentice Hall! ===========================================================================