Re: Tuning advice.

Madison Pruet wrote: > > Well, you do need more buffers. Generally the read cache rate needs to be > pretty close to 99%. You might be able to reduce your checkpoint times even > more by setting LRU_MAX_DIRTY to 2. > > You also have skewed usage of the chunks. From your sample ol_ch_mars_a is > accounting for a rather large percentage of your activity. You might want to > consider moving some of the tables in there to othere chunks. This might help > your checkpoint times also. > > I would consider setting DD_HASHSIZE to a much larger value. I don't know where > all of your IO is coming from, but if it is dictionary information, this would > help. If oninit is only taking 50% of the system when it is overloaded, that > implies that you are blocking quite a bit on IO completion. Increasing buffers > would help. > > I'm sure that some other folks that frequent the news group will have some other > ideas and/or suggestions. I'll be interested to see what they suggest. I like your suggestions myself. Art S. Kagel > Tony Flaherty wrote: > > > HPUX 10.20 > > > > Dual processor K200 with 512Mb Ram > > > > IDS 7.30.UC7 > > > > OLTP system running apps and instance on same box using shared memeory > > connection, approx. 100 users. The top output can show oninit running as > > high as 90% when one of our "poor" applications is running (3rd party > > software s%&t) but generally this is <50% > > > > Earlier this year I increased the number of buffers allocated, this appears > > to have resulted in an increase in checkpoint times from an average of 0 > > seconds to an average of 1 second with frequent occurances of 2,3,4 and > > higher. There are some very high peaks of 15+ but these are down to > > problems with some applications and users running big updates during the > > day. > > > > As it happens the latest Tech Notes has an article on tuning checkpoints and > > this mentions using onstat -g ioq to monitor the AIO VPS (output below) I > > always thought I needed approx one per chunk of these and I'm already > > running with more than that, should I look at increasing this? > > > > I also think I may be running into some disk related bottlenecks. > > > > Some onstat info and the onconfig follow. [SNIP]