Re: High numbers of buffer waits in OnLine

David Williams (djw@smooth1.demon.co.uk) wrote: : In article <31DA19B7.4A6F@west.co.za>, Mark D Stock <marks@west.co.za> : writes : >David Williams wrote: : ><SNIPedySNIP> : >> First apologies for the typo above - f and g are next to one another : >> on the keyboard! : >> : >> Buffer wait does mean that the process is waiting for a buffer which : >> is in use by another process. However they do not have to be accessing : >> the same data. What if you want to read in a page from disk and all : >> the buffers are already in use? i.e. the page is not already in the : >> buffers but all the buffers are in use so you have to wait for a : >> buffer to become free. : > : >This would show up as foreground writes. Which are VERY undesirable! : > : No if processes are use the buffers to read you cannot free the : buffer until the process has read the buffer it requested. Sorry for answering late (and sorry for my English too), but I was out for several days. You are right with your buffer wait description, but: if the process waits for another process in situation as you mentioned above, IT IS because the process need to execute foreground write! So the primary problem is foreground writes, not buffer waits. I think this would increase OnLine statistics in buffer waits and also foreground writes. If buffer waits occure without fg writes, it would be situation I described in my previous post and I think that it is not a real problem. : The Informix person whjo gave the course checked with the guys the : States who had access to the source code to OnLine. : When I increased the number of buffers on a clients machine which had : many buffer waits ( it was a 10 CPU Pyramid machine with 128 users, : 512Mb RAM / 10 Gb disk) this gave a noticable increase in the : performance. : ! : -- : David Williams -- ================================= Miroslav Lorenc Software Applications and Systems E-mail: Miroslav.Lorenc@sasprg.cz Prague Czech Republic