History cache performance?

[bill davidsen]

In article <836069667.20010307103629 at rrz.uni-koeln.de>,
Bernhard Kohl  <B.Kohl at rrz.uni-koeln.de> wrote:
| 
| Hello inn-workers,
| 
| [1]  does anybody know "reasonable" values for history cache? I didn't
|      find anything in the FAQs. Does the following table point out to a
|      problem with the history files?
| 
|           Reason          Count           %Count
|           -------------   ------          ------
|           Positive hits   950874          37.8%
|           Negative hits   880000          35.0%
|           Do not exist    408311          16.2%
|           Cache misses    275792          11.0%
|           -------------   ------          ------
|           TOTAL: 4        2514977         100.0%

  I think your miss percentage is high. I usually increase the size
until I get below 2% or until I get no improvement by going larger. If
you have slow feeds you are not going to easily reduce the value.

  You didn't say much about your feed, to some extent it depends on
incoming article rate. Try 200-500k times arts/sec (ie. 4-10MB for 20
arts/sec) as a starting point if you like. How much time are you
spending in HIShave? If it's small, then you have very little possible
gain and the misses are not hurting you a lot.

  Voodo, all voodoo...

  BTW: anyone tried playing with locking the cache in memory? I tried
using a locked shared mem, but I didn't see a benefit. My machine had
lots of memory, though. I would expect the cache to stay in memory on a
loaded machine, but comments here made me try it.
-- 
bill davidsen <davidsen at tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.