Re: Performance Problems with 7.30.UC7XB

Hi Doug, I have read the replies so far and will try not to repeat where I cannot add anything so if I do not comment on something someone else has don't ignore the advice, it all looked good to me. Doug Agnew wrote: > > Running Informix 7.30.UC7XB (for SAP) on > HP-UX 10.20 -- machine is T520 w/4 processors and 1.75 Gb memory. EMC > Symmetrix with HP-PB Fast/Wide SCSI connections. > > Experiencing cache-hit read % of < 80% (and needless to say, our users are > screaming) [SNIP] > onstat -p shows (a 6.5 minute sample) [SNIP]> ixda-RA idx-RA da-RA RA-pgsused lchwaits > 648128 251 4837 386873 57477 This shows a readahead utilization ratio of <60%. One problem is excessive readahead is flushing the buffer cache and reducing the read cache%! I'll look at the parameters below. > environment variables that have been set (per info from SAP) > NO_KEYFIRST=1 > NO_SUBQF=1 > SINGLESEG_OFF=1 > DBBLOBBUF=64 > > onconfig file follows: > # > # C H A N G E . L O G > # --------------------- > # > # rag-10/10/97 chg'd NUMCPUVPS from 2 to 4. > # > # rag-10/10/97 chg'd cause we added 768 mb mem to server. > # is wuz > # SHMVIRTSIZE 870000 450000 > # BUFFERS 300000 210000 > # > # rag-10/10/97 chg'd per SAP Carl Rossini & Sari Nathans - Golive > Perf.problem > # > # LRU_MAX_DIRTY 2 5 > # LRU_MIN_DIRTY 1 3 > # CLEANERS 30 40 > # CKPTINTVL 600 900 (from 15 minutes to 10 minutes). > # NETTYPE soctcp,1,120,CPU per Sari Nathans > # > # rag-10/09/97 chg'd per SAP Dave Green for SAP Golive performance problem. > # > # MULTIPROCESSOR 0 (cpu) 0=1 cpu, 1=2 cpu > # NUMCPUVPS 1 (cpu) 1=1 cpu, 2=2 cpu > # SINGLE_CPU_VP 1 (cpu) 1=1 cpu, 0=2 cpu > # > # rag-07/29/97 made following chenges per Sari Nathans > # and EarlyWatch Subhash per Sari Nathans > # new old > # --- --- > # LRUS 64 32 > # NETTYPE soctcp,1,60,CPU soctcp,1,100,NET (and EarlyWatch) > # RA_PAGES 32 8 > # RA_THRESHOLD 30 4 > # > # > # rag-06/04/97 NUMAIOVPS 10 -> 16 # Number of AIO vps. note 41360 > # > # mlc-03/03/97 MULTIPROCESSOR 1 (from 0) > # NUMCPUVPS 2 (from 1) > # SINGLE_CPU_VP 0 (from 1) > # BUFFERS 210000 (from 70,000) > # SHMVIRTSIZE 250000 (from 132,000) > # > #************************************************************************** > # Root Dbspace Configuration > > ROOTNAME rootdbs # Root dbspace name> ROOTPATH /informix/PRD/sapdata/physdev1/data1 > # Path for device containing root dbspace > ROOTOFFSET 0 # Offset of root dbspace into device > (Kbytes) > ROOTSIZE 600000 # Size of root dbspace (Kbytes)> > # Disk Mirroring Configuration Parameters > > MIRROR 0 # Mirroring flag (Yes = 1, No = 0) > MIRRORPATH /informix/PRD/sapdata/physdev2/data2 > # Path for device containing mirrored root > MIRROROFFSET 0 # Offset into mirrored device (Kbytes)> > # Physical Log Configuration > > PHYSDBS physdbs # Location (dbspace) of physical log > PHYSFILE 298894 # Physical log file size (Kbytes)> > # Logical Log Configuration > > LOGFILES 200 # Number of logical log files > LOGSIZE 10000 # Logical log size (Kbytes) rag-10/25/96> > # Message Files > > MSGPATH /informix/PRD/online.chpfdb1.prd.log > # System message log file path > CONSOLE /informix/PRD/console.chpfdb1.prd.log > # System console message path > > # System Archive Tape Device > > TAPEDEV /informix/tape # Tape device path rag-01/24/99 > TAPESIZE 90000000 # rag-11/12/98> > ##TAPEDEV /dev/tape120 # Tape device path rag-03/01/97 > ##TAPESIZE 6000000 # rag-04/10/97 > > TAPEBLK 256 # Tape block size (Kbytes)> > # Log Archive Tape Device > > LTAPEDEV /dev/tape120 # Log tape device path > LTAPEBLK 256 # Log tape block size (Kbytes) > LTAPESIZE 7000000 # Max amount of data to put on log tape > (Kbytes)> #LTAPEDEV /saplog/PRD # Log tape device path > #LTAPEBLK 2048 # Log tape block size (Kbytes) > #LTAPESIZE 4000000 # Max amount of data to put on log tape > (Kbytes) > > # Optical > > STAGEBLOB ,1 # INFORMIX-OnLine/Optical staging area > > # System Configuration > > SERVERNUM 0 # rag-02/08/96 id corresponding to a OnLine> instance > DBSERVERNAME chpfdb1prdshm # Name of default database server > DBSERVERALIASES chpfdb1prdtcp # List of alternate dbservernames > NETTYPE ipcshm,1,60,CPU # Override sqlhosts nettype parameters > NETTYPE soctcp,1,120,CPU # Override sqlhosts nettype Network listeners should NOT be in the CPU VP. The CPU cannot block on the network port so it has to poll in between the other things it has to do like execute queries. Listening reduces throughput. Change the soctcp NETTYPE to NET VP. > DEADLOCK_TIMEOUT 60 # Max time to wait of lock in distributed > env. > RESIDENT 0 # Forced residency flag (Yes = 1, No = As someone already mentioned residency of the BUFFER CACHE is critical to performance. > > # The following (3) parms relate to the number of CPU processors. > # --------------------------------------------------------------- > MULTIPROCESSOR 1 #(cpu) 0=1 cpu, 1=2 cpu. rag-10/11/97 > NUMCPUVPS 4 #(cpu) 1=1 cpu, 2=2 cpu. rag-10/18/97 On a fast machine like this you may even be able to configure up to 2X the number of CPUs for NUMCPUVPS. Experiment. > SINGLE_CPU_VP 0 #(cpu) 1=1 cpu, 0=2 cpu. rag-10/11/97> > MUTEX_WAIT_LISTS 0 # If non-zero, force use of mutex wait > lists > > NOAGE 0 # Process aging HP/UX is aggressive about aging long running process priority so turning NOAGE on is VERY important. The main downside is if you have any other long running processes they will be starved faster with the IDS engine NOAGEd if they are not as well. > AFF_SPROC 0 # Affinity start processor rag-08/25/96 > AFF_NPROCS 0 # Affinity number of processors rag-08/25/96 Affinity seems to help on HP systems according to most posts from HP users. > # Shared Memory Parameters > > #LOCKS 768000 > LOCKS 1500000 > BUFFERS 210000 # Max shared buffers rag-06/12/1999 > NUMAIOVPS 60 # Number of AIO vps

Art, Thanks very much for the input. Now I just have to go twist some tech-support arms. We have KAIO disabled currently, and it will be a big fight to enable it <g>. Doug Art S. Kagel wrote in message <377A2D92.69CEAEEC@bloomberg.net>... >Hi Doug, > I have read the replies so far and will try not to repeat where >I cannot add anything so if I do not comment on something someone else >has don't ignore the advice, it all looked good to me. > <excellent suggestions snipped> >Art S. Kagel