Re: BIG Istar performance problem

Paul, thanks for fast response, >> On the server I have 300 to 500 sqlturbo processes. No frontends >> runs on the server. All frontends are version 4.10.xx and all >> backends are version 5.00.xx. The slow connection with Istar is done by tcp/ip. >We have a Sun IPC (small, single cpu desktop) connected to an RS/6000 model >550 via a 56K WAN circuit. We are using 4.10.x tools and 4.10.x >engine(s). >The RS/6000 has 50-60 users generating 200+ sqlturbo processes. The Sun >has under 12 users and only 25 or so sqlturbo processes. Neither has the >same horsepower as your boxes! >> If I run my test program on the server, when the load is high >> without using Istar, the total used time is very low, it may >> be 0.07 sec. up to 1.00 sec. >Local processing is on the order of 1-3 seconds, I-Star processing is on >the order of 3-6 seconds. I would give a rough estimate of I-Star (and >data circuit) overhead of a factor of two. On a Pyramid MIServer I have seen an increase of up to 2 sec. with Istar. >I am not sure what factors are increasing your processing time by a factor >of 190X!! Yes that was my peak, more normal is 100X, but that does not make it better ! >I should add that both engines are tuned very well. The RS/6000 has about >24 Meg of shared memory dedicated to On-Line and our table sizes are pretty >nominal (10K to 15K records). I think so, last time i made a tbstat -p, it showed 99.5 % read cache and 98 % write cache, the write cache may fall to 92 - 93 % and fast checkpoints (low LRU marks) on the main OnLine. Total shared memory is 38 Mbytes. Thanks ----------------------------------------------------------------- Databaseadministrator Poul Pedersen Kuwait Petroleum (Danmark) A/S Hummeltoftevej 49 Mail: pp@q8.dk 2830 Virum Voice: +45 45 98 45 94 Denmark Fax: +45 42 85 14 18 -----------------------------------------------------------------