Re: 0% caching

>From: uunet!enuucp.eas.asu.edu!anasaz!qip.naomi (Naomi Walker) >Subject: 0% caching >Date: Thu, 12 Nov 92 16:28:24 MST >X-Informix-List-Id: <list.1607> >Here's one that has me stumped. We are running Online, on an AST 486 >with Interactive Unix. We have been tuning tbconfig to help some performance >problems, and last experiment showed something odd. When we changed BUFFERS >from 512 to 2048, % cached (disk writes), flopped to 0%. I've never seen >such a thing. Can anyone offer a suggestion why buffered write would be >smaller than disk writes? >BIGreads dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached >57 12143 14763 692592 98.25 9076 18262 7884 0.00 This is perfectly reasonable: you have 8000 writes to 700000 reads, so your database is doing almost no update work n relation to the amount of queries, so the write cache efficiency is irrelevant. I have a client whose OnLine is taken offline and then brought back online every Sunday. From then until Wednesday am (the wee small hours), the write cache efficiency is 0%, and the ratio of reads to writes is, if anything, more extreme. On Wednesday morning, a major batch update occurs, and the write cache ration is restored to a normal value, not that the number of writes changes very much for the rest of the opeational week. The number of bufwrits being smaller than the number of pagwits simply means that you are frequently writing more than one page in a buffer; at least, that's how I'd interpret it. Yours, Jonathan Leffler (johnl@obelix.informix.com) #include <disclaimer.h>