Re: Problem with SOC vp sys cpu usage sometimes getting high.

Posted in 2001

Topics: Server Administration, Networking & sqlhosts Configuration, Platform-Specific Issues

Frank Langelage — Sat, 03 Feb 2001 20:35:49 +0100 — source: Informix-list mailing list archive (1991-1998)

kaien@neowiz.com wrote: > > I had IDS2000 on AIX 4.3.3, S/80, 12 cpu. > I had a problem with cpu utilization. > > This is SAR output of normal situation. > > 01:41:20 %usr %sys %wio %idle > 01:41:21 29 14 0 57 > 01:41:22 33 17 0 50 > 01:41:23 34 14 0 52> > But sometimes, System fall into situation like this. > > 01:45:28 %usr %sys %wio %idle > 01:45:29 38 29 0 33 > 01:45:30 33 25 0 42 > 01:45:31 35 31 0 34 > [ snip ] > ONCONFIG > > .... > SERVERNUM 0 # Unique id corresponding to a OnLine instance > DBSERVERNAME neowiz_shm # Name of default database server > DBSERVERALIASES neowiz,ndb_cpool1,ndb_cpool2,ndb_cpool3,ndb_cpool4,ndb_cpool5,ndb_cpool6 > # List of alternate dbservernames > > NETTYPE ipcshm,1,200,CPU > NETTYPE soctcp,15,512,NET 15 Informix processes for Socket communication with 512 Users each ? How many network users does the engine have to serve at max ? I think you should lower the number 512 to 250 or 200. If you can, lower number proc. from 15 to ... ? But why are there 20 soc-vp ? > CLEANERS 32 # Number of buffer cleaner processes > LRUS 128 # Number of LRU queues > LRU_MAX_DIRTY 4 # LRU percent dirty begin cleaning limit. UPDATE > LRU_MIN_DIRTY 2 # LRU percent dirty end cleaning limit. UPDATE You should increase CLEANERS up to LRUS = 128. Regards Frank

↪ replying to Frank Langelage

kaien@neowiz.com — Sun, 4 Feb 2001 04:54:24 +0900 (KST) — source: Informix-list mailing list archive (1991-1998)

Thanks for your concern. Frank Langelage <frank@lafr.de> wrote: > kaien@neowiz.com wrote: >> >> I had IDS2000 on AIX 4.3.3, S/80, 12 cpu. >> I had a problem with cpu utilization. >> >> This is SAR output of normal situation. >> >> 01:41:20 %usr %sys %wio %idle >> 01:41:21 29 14 0 57 >> 01:41:22 33 17 0 50 >> 01:41:23 34 14 0 52>> >> But sometimes, System fall into situation like this. >> >> 01:45:28 %usr %sys %wio %idle >> 01:45:29 38 29 0 33 >> 01:45:30 33 25 0 42 >> 01:45:31 35 31 0 34 >> [ snip ] >> ONCONFIG >> >> .... >> SERVERNUM 0 # Unique id corresponding to a OnLine instance >> DBSERVERNAME neowiz_shm # Name of default database server >> DBSERVERALIASES neowiz,ndb_cpool1,ndb_cpool2,ndb_cpool3,ndb_cpool4,ndb_cpool5,ndb_cpool6 >> # List of alternate dbservernames >> >> NETTYPE ipcshm,1,200,CPU >> NETTYPE soctcp,15,512,NET > 15 Informix processes for Socket communication with 512 Users each ? > How many network users does the engine have to serve at max ? > I think you should lower the number 512 to 250 or 200. > If you can, lower number proc. from 15 to ... ? > But why are there 20 soc-vp ? I'm running web service. About 80k concurrent users peak. But Informix has about 1000 sessions. And it is fixed. I'm using some connection pooling 3-tier and web servers are not directly request to informix.(So number of database session is independent to real users.) I've tried various number of socket thread and connections per thread, from 10, 200 to 40,512 , But didnt help and phenomenon was same. >> CLEANERS 32 # Number of buffer cleaner processes >> LRUS 128 # Number of LRU queues >> LRU_MAX_DIRTY 4 # LRU percent dirty begin cleaning limit. UPDATE >> LRU_MIN_DIRTY 2 # LRU percent dirty end cleaning limit. UPDATE > You should increase CLEANERS up to LRUS = 128. Isn't it to much? (CLEARERS 32 is refer to number of dbspaces I have.) > Regards > Frank

Glossary

onconfig