I agree that 120 hard page faults per second sounds high, so you may be using more memory than you've got. Have a look at the page outs (which you can get from nsar -g). If they are running at only a few per second, then the page faults relate to named pages and you need to tune the operating system. If so, read the chapter on memory tuning in Analyzing DG/UX System Performance for help. Otherwise, you should probably tune down Oracle's memory usage, or buy some more memory.

I was going through our performance tuning guide for DG/UX (pretty sparse) and was running some of the samples and saw one that kind of concerned me. The guide said that having over 15 memory page faults/sec was poor performance. I'm seeing numbers avg about 120 faults/sec. I'm just wondering if this is normal for a large data warehouse where a lot of full table scans over large tables are being performed or if we really have a problem here. The numbers, just during the short time I was looking at them were up as high as 300 at one point. Any good general OS tuning books for Unix? You can't find anything regarding DG/UX and their documentation is pretty poor as far as I can see.