Slow responses from remote application server

Hoping somebody might have some ideas regarding some response issues we are having. Let me set the scene. Database (IDS10) running on a Sun Sparc M5000 server with 8 Quad Core CPUs and 32G of memory. Approx 2000 concurrent sessions. Some of the key config params are: - NUMCPUVPS 20 (Will probably set this higher shortly) NETTYPE tlitcp,6,3000,NET (set to 3000 as did not realise this was per thread) NUMAIOVPS 16 BUFFERPOOL size=2K,buffers=8000000,lrus=128,lru_min_dirty=0.025000,lru_max_dirty=0.100000 RESIDENT 1 SINGLE_CPU_VP 0 If we run the application in a container on the same physical server as the DB our test query consistently takes 3-4 seconds to complete. If we run the application in a container on a DIFFERENT server (Sun v490) the same query consistently takes about 30 seconds to complete. Massive difference... These servers are connected via a Gigabit Ethernet connection with a single switch in between. The network latency does not appear too bad: - 26 packets transmitted, 26 packets received, 0% packet loss round-trip (ms) min/avg/max/stddev = 0.207/0.5816/1.45/0.366 We have not implemented the change yet, but we believe we will get some improvements by using OPTOFC. We are also considering setting the FASTPOLL config parameter. Does anybody have any ideas on how to improve these response times? Thanks Paul Ridding Tourism Technology.

On 28.04.10 07:19, PAUL RIDDING wrote: > Hoping somebody might have some ideas regarding some response issues we are > having. Let me set the scene. > > Database (IDS10) running on a Sun Sparc M5000 server with 8 Quad Core CPUs and > 32G of memory. > > Approx 2000 concurrent sessions. Some of the key config params are: - > > NUMCPUVPS 20 (Will probably set this higher shortly) > NETTYPE tlitcp,6,3000,NET (set to 3000 as did not realise this was per thread) > NUMAIOVPS 16 > BUFFERPOOL> size=2K,buffers=8000000,lrus=128,lru_min_dirty=0.025000,lru_max_dirty=0.100000 > RESIDENT 1 > SINGLE_CPU_VP 0> > If we run the application in a container on the same physical server as the DB > our test query consistently takes 3-4 seconds to complete. > > If we run the application in a container on a DIFFERENT server (Sun v490) the > same query consistently takes about 30 seconds to complete. Massive > difference... These servers are connected via a Gigabit Ethernet connection > with a single switch in between. The network latency does not appear too bad: > - > > 26 packets transmitted, 26 packets received, 0% packet loss > round-trip (ms) min/avg/max/stddev = 0.207/0.5816/1.45/0.366 > > We have not implemented the change yet, but we believe we will get some > improvements by using OPTOFC. We are also considering setting the FASTPOLL > config parameter. > > What kind of statements are executed from the application? One query which retrieves a big list of data for the client or a lot of small queries (single row selects, inserts, etc). If it's more like the second option OPTOFC might help a little bit. Our application is like this and we also see a significant performance drop if application and database are on different servers, even with a dedicated 10 GB ethernet between them.

Six poll theads isn't quite enough for 2000 connections and certainly not i= f you expect a peak of 1.5 times that. Try this: NETTYPE tlitcp,10,300,NET IBM's rule of thumb is that a single poll thread can handle ~350 connection= s but <300 is best performance. FASTPOLL 1 definitely a help. Art=20 -----Original Message----- From: PAUL RIDDING <pridding@tt.com.au> Sent: Wednesday, April 28, 2010 12:19 AM To: ids@iiug.org Subject: Slow responses from remote application server [19907] Hoping somebody might have some ideas regarding some response issues we are= =20 having. Let me set the scene.=20 Database (IDS10) running on a Sun Sparc M5000 server with 8 Quad Core CPUs = and=20 32G of memory.=20 Approx 2000 concurrent sessions. Some of the key config params are: -=20 NUMCPUVPS 20 (Will probably set this higher shortly)=20 NETTYPE tlitcp,6,3000,NET (set to 3000 as did not realise this was per thre= ad)=20 NUMAIOVPS 16=20BUFFERPOOL=20 size=3D2K,buffers=3D8000000,lrus=3D128,lru_min_dirty=3D0.025000,lru_max_dir= ty=3D0.100000=20 RESIDENT 1=20 SINGLE_CPU_VP 0=20 If we run the application in a container on the same physical server as the= DB=20 our test query consistently takes 3-4 seconds to complete.=20 If we run the application in a container on a DIFFERENT server (Sun v490) t= he=20 same query consistently takes about 30 seconds to complete. Massive=20 difference... These servers are connected via a Gigabit Ethernet connection= =20 with a single switch in between. The network latency does not appear too ba= d:=20 -=20 26 packets transmitted, 26 packets received, 0% packet loss=20 round-trip (ms) min/avg/max/stddev =3D 0.207/0.5816/1.45/0.366=20 We have not implemented the change yet, but we believe we will get some=20 improvements by using OPTOFC. We are also considering setting the FASTPOLL= =20 config parameter.=20 Does anybody have any ideas on how to improve these response times?=20 Thanks=20 Paul Ridding=20 Tourism Technology.=20 ***************************************************************************= ****=20 Forum Note: Use "Reply" to post a response in the discussion forum.=20

Thanks Art. I have made the suggested changes and will bounce the engine this weekend. Regards, Paul