Re: Outrageously low write cache %

I don't have any statistics before configuring ER and setting logging to unbuffered but I'm pretty sure that buffer writes % weren't that LOW. Right now I'm looking at the logical log entries to see what is being written. I'll post whatever I find here ... As always, thanks for your reply Madison. In article <3A636984.7163D332@home.com>, Madison Pruet <mpruet@home.com> wrote: > Do your buff reads normally run about 200X the bufwrites? Do you have some > statistics prior to making your changes? > Sent via Deja.com http://www.deja.com/

OK - let's understand exactly what the write cache rate really means. As you are aware, we strongly encourage sites to get their read cache rate really high (i.e. above 98%), but that is because 1) most systems do far more reads than writes and 2) a lot of the reads should be against index pages, the catalogue pages, and system pages such as the partition and bitmap pages. However, for the write cache rate, we generally don't want to see too high of a number unless the server is doing a lot of updates and we want to avoid having to read before write. If the server is only receiving inserts, then I would not expect to see a high rate, especially if those inserts don't have indexes. For instance if the table receiving inserts only could contain two physical rows, then it would be impossible for the write cache rate to be higher than 50%. Also if the LRU levels are set really low, then I would expect the write cache rate to be low. Why? - Because by setting the LRU levels low, the dirty page will be more likly to be flushed in between checkpoints and if the page is flushed, then it is no longer dirty and is more likely to be removed from the buffer pool. And then if a subsequent write needs to be done on that page it is not cached. So basically, before we get too worried about the write cache rate, it might be best to get a baseline of what is normal for the instance and what the characteristics of the machine are. (i.e. what kind of activity is being done on the machine.). From the looks of it, there is a huge difference between the buffer reads and the buffer writes. Again that means that the write cache rate is going to be low, simply because the ratio is so low. Lyzander Marantal wrote: > I don't have any statistics before configuring ER and setting logging > to unbuffered but I'm pretty sure that buffer writes % weren't that LOW. > > Right now I'm looking at the logical log entries to see what is being > written. I'll post whatever I find here ... > > As always, thanks for your reply Madison. > > In article <3A636984.7163D332@home.com>, > Madison Pruet <mpruet@home.com> wrote: > > Do your buff reads normally run about 200X the bufwrites? Do you > have some > > statistics prior to making your changes? > > > > Sent via Deja.com > http://www.deja.com/