Re: ESQL/C library functions - Any change from 5.x to 7.x
Posted in 1997
On Wed, 8 Oct 1997, Richard Spitz wrote: > Art S. Kagel wrote: > > Anyway, I have never found the relay module to be a bottle neck > > compared to 5.0x native access at all. > It occurs to me that I did not answer you specific questions in my previous reply, so. > How does the overall performance of your engine compare to the > previous 5.x version? Are you using the same hardware for the > 7.x engine that you used for the 5.x engine? > > I can imagine that 7.x on faster (multiprocessor?) hardware outper- > forms the previous engine by so much that the performance hit delivered > by using the relay module is more or less neutralized. This I think I have covered, in general. Specifically, we used to have to run from 2 to 5 instances of OL5.0x to handle the data and user load (some databases were too big for a single instance others fit but could not handle the load if they shared the instance with another database). Although 5.0x shared the machine well and the multiple instances mostly ran well the single 7.21 instance we now use is far faster. Same CPUs, though in the process of upgrading we have increased from 16CPUs to 24 or 32CPUs and changed from RAID5 to stripes and mirrors. However, when we were evaluating the 7.xx product we were using the exact same hardware as was being used for 5.0x and production type loads. Since most of our queries are simple (1-2 tables indexed searches non-fragmented tables except where size or performance required fragmenting and PDQPRIORITY=0) we are not taking best advantage of all of those CPUs, most queries use only one SQLEXEC thread and it is sharing one of 28 CPU VPs with 120 other processes which are application and data servers for 80,000 users. You decide if my results are valid for you. I repeat that whenever we have had to use the Relay Module, as a short term solution, because we were in a rush to drop and recycle the 5.0x instances or because we were spending our resources on rewriting the application to take better advantage of 7.xx features, we noticed not a decrease in performance over the application running against 5.0x but a slight improvement. Believe me I do not advocate not recompiling. On the contrary, I advocate doing that so vigorously, in combination with rewriting to use new features, that you will sometimes use the Relay Module as a quick way to get rid of 5.0x while the process continues. I suppose if I had said 'Your mileage may vary' I would have gotten less resistance to the idea. Art S. Kagel, kagel@bloomberg.com