Re: poll threads

Bill, We have recently gone through (and still experience some connection problems) the same problem. In our environment, we had the following settings: DATABASESERVER sparky_shm DATABASERVERALIAS sparky_tcp NETTYPE tlitcp,4,500,CPU # This number is high and is getting lowered. NUMCPUVPS 3 For the longest period of time (2-3 months), I thought that we running on a CPU vp class. But in fact, we were running on a NET vp class. ODS determined that more CPU VPs were being requested (NETTYPE) than were defined (NUMCPUVPS). ODS defaults to running on NET vp class in this case. Two weeks ago, I saw a message in our log file and made this change NETTYPE tlitcp,3,500,CPU. This is when we noticed very bad response time. The reason for the 180 degree turn around in performance was due to the fact that the poll threads were getting swapped in/out. Where the poll threads run as NET vp class you don't have this same behavoir. I chose CPU vp class becaue the manual suggests putting high-traffic polls on CPU. There was a one-line in the Performance an Tuning manual (from class) on pg. 9-189 that mentions, "oh by the way, you might want to use NET when a lot of remote users connect". Output of onstat -g act @sparky:/opt/informix $ act INFORMIX-OnLine Version 7.23.UC1 -- On-Line -- Up 3 days 16:13:30 -- 388496 Kb ytes Running threads: tid tcb rstcb prty status vp-class name 7 180b4e50 0 2 running 9shm sm_poll 8 180c8b18 0 2 running 10tli tlitcppoll 9 180c9058 0 2 running 11tli tlitcppoll 10 180c95c8 0 2 running 12tli tlitcppoll 11 180c9b38 0 2 running 13tli tlispxpoll 19541 191c0a28 1b0cb3d8 2 running 1cpu sqlexec 22550 18488808 1cc7d798 2 running 4cpu sqlexec 22697 1b0928e8 1cc62618 2 running 3cpu sqlexec Notice that sm_poll, tlitcppoll, and tlispxpoll are running as NET vp classes. If they were running as CPU vp class, you would see something like 1cpu. When are tlitcppoll threads were running as CPU vp class, they were continually swapped in/out. Run onstat -g act -r 1 to watch behavior. Also use onstat -g sch to vp classes. You asked "how many poll threads?". Don't know. We are experimenting with that know. Good luck... SteveR Bill Weaver wrote: > What's the best way to monitor poll threads to determine if you need additional > ones? We have been having a problem with sessions taking a while to actually make > a database connection. When additional poll threads were added, this problem > seemed to go away. My question is what should I be looking at to determine if > additional poll threads are needed or not needed? We are not hitting the maximum > number of connections a poll thread can service as defined in NETTYPE, but it does > appear that the ones we do have may be overloaded. Where do I look to see if that > is what is happening? > -------------------------------------------------------- > Name: Bill Weaver > E-mail: Bill Weaver <billw@fscorp.com> > Date: 04/14/98 > Time: 13:26:40 > > This message was sent by Z-Mail Pro - from NetManage > NetManage - delivers Standards Based IntraNet Solutions > -------------------------------------------------------- -- ----------------------------------------------------------------- Steve Romankiw + Executive Risk Inc. + DBA + email: sromankiw@execrisk.com 82 Hopmeadow Street + work: (860) 408-2474 Simsbury,CT 06070 + fax: (860) 408-2139 -----------------------------------------------------------------