RE: You've heard it all before: newer, faster server runs slower than older, slower server after data migration

Sorry about that, UPDATE STATS was run (I even used Art's program to do it). Another thing I would like to add. VP Scheduler Statistics: vp pid class semops busy waits spins/wait 1 28505 cpu 17491 17534 1000 2 28506 adm 0 0 0 3 28507 cpu 20517 20586 999 4 28508 cpu 19405 19467 999 5 28509 cpu 18285 18334 999 6 28510 cpu 69192125 69325060 999 7 28511 cpu 56643372 56765184 999 8 28512 cpu 44975001 45084655 999 9 28513 lio 2 0 0 10 28514 pio 2 0 0 11 28515 aio 146 0 0 12 28516 msc 30523 0 0 There are a lot of semops and busy waits on VP's 6,7,8. Is this ok since the spins/wait are at or near 1000? Thanks Andrew -----Original Message----- From: Jonathan Leffler [SMTP:jleffler@informix.com] Sent: Monday, November 20, 2000 3:32 PM To: Andrew Ford Subject: Re: You've heard it all before: newer, faster server runs slower than older, slower server after data migration Andrew Ford wrote: > > I've seen this problem many times in the news groups, and each time I've > said to myself, "That poor soul, someone spent a load of money for a new > server expecting a performance increase, only to find that the new server > runs dog slow. Thank God I don't have that problem." You don't mention UPDATE STATISTICS in the first few paragraphs? Don't forget Art Kagel's program for generating UPDATE STATISTICS from the utils2_ak package in the IIUG archive (http://www.iiug.org). > Help, I have that problem. > > The basics, onconfig and some onstat output at end of message: > HP-UX billsvr1 B.11.00 U 9000/800 > Informix Dynamic Server Version 7.31.FC4 > 8 Processors > 4 Gig Ram > OLTP/DSS combo, OLTP more important than DSS. > > What we did: > We moved all Informix processing from an 8 Processor, 4 Gig Ram N Class > HP/UX machine to a 8 Processor, 4 Gig Ram N Class HP/UX machine. > The new configuration is connected to an XP256 with roughly 55 47 Gig > Drives and 11 15 Gig Drives (usable in a mirrored pair config) split into > 1305 Informix Chunks which make up 454 Dbspaces. > We started up ER in a Primary/Target config (the large N Class is the > server, replicating to 7 smaller targets), we initially used Unbuffered > logging (the old system used No Logging) but I recently set the logging to > Buffered to help speed things up a bit. > Most of our code has been modified to use Isolation Dirty Read to cut down > on lock contention. > > What I think the problem is, but don't know how to stop it: > I found the system CPU time (via top) was constantly at 20% - 30%, and sar > -w 3 10 shows (what I think) is a large number of context switches. > > HP-UX billsvr1 B.11.00 U 9000/800 11/20/00 > > 11:50:42 swpin/s bswin/s swpot/s bswot/s pswch/s > 11:50:45 0.00 0.0 0.00 0.0 30094 > 11:50:48 0.00 0.0 0.00 0.0 27198 > 11:50:51 0.00 0.0 0.00 0.0 24618 > 11:50:54 0.00 0.0 0.00 0.0 28735 > 11:50:57 0.00 0.0 0.00 0.0 31980 > 11:51:00 0.00 0.0 0.00 0.0 27254 > 11:51:03 0.00 0.0 0.00 0.0 25524 > 11:51:06 0.00 0.0 0.00 0.0 30420 > 11:51:09 0.00 0.0 0.00 0.0 31634 > 11:51:12 0.00 0.0 0.00 0.0 34723> > Average 0.00 0.0 0.00 0.0 29218 > > I had initially configured the tcp threads to run on 6 NET VP's (NETTYPE > soctcp,6,35,NET), but when I noticed the context switching going on I > configured them to run on CPU VPs (NETTYPE soctcp,3,80,CPU) but this did > not help. > > Basically I'm stumped. I have tried to tune Informix but I beleive the > problem lies within HP/UX kernel config. > > Any help that anyone can offer will, as always, be greatly appreciated. > > Thank You, > > Andrew Ford -- Yours, Jonathan Leffler (Jonathan.Leffler@Informix.com) #include <disclaimer.h> Guardian of DBD::Informix v1.00.PC1 -- http://www.perl.com/CPAN "I don't suffer from insanity; I enjoy every minute of it!"