Hi all, I'm analyzing some performance data that was gathered during a period of one week and am clueless about some results. Here's the context: Server: IBM xSeries 360, 4x Xeon MP 2.0GHz, 2GB PC1600 SDRAM, IBM 36,4GB SCSI HD OS: Windows 2000 Server Running: Oracle 9iAS (Oracle application server) Average load: 50 concurrent users Average CPU usage is low at 4%, average available memory is high at 1,2GB, average paging file usage is low at 4% and average disk usage is low at under 1%. Everything seems normal, except for this: average page faults per second is over 100. I have run similar data collection on the database server (approximately the same configuration). The results are pretty much the same in terms of CPU, memory, paging file and disk usage. The page fault count is even higher at over 700! I have looked at the graph and there are page fault spikes every couple of seconds. I have been told that the accepted threshold for page faults is around 20. These results are far higher than this. Is this normal? How is this affecting system performance? What could be causing this? Thanks a lot, GB

The only thing that I can think of is that there are short periods where the free memory goes to zero and there is thrashing with many thousands of page faults. That could raise the average. I think I would try and get a graph of time and page faults.

Yes, that seems to match the actual situation. Page faults average 10-15 per second for a couple (~10) of seconds and then there's a spike of 50 or more. Correct me if I'm wrong, but this would happen only if memory was running low, which is definitely not the case here: available memory is over 1GB! What else could cause this? Thanks!

Since this is a weekly average, it could be running fine for 6.99 days. Then at some point something may be causing all the memory to be used up for maybe a few minutes and causing many thousands of page faults. This probably would have very little affect on the free memory statistic, but could have a large affect on the page fault statistic. A time/page fault graph would tell for sure.