how does history file get too large?

bill davidsen davidsen at tmr.com
Thu Jan 29 21:31:18 UTC 2004


In article <87ptd3hjcb.fsf at windlord.stanford.edu>,
Russ Allbery  <rra at stanford.edu> wrote:
| Anne Wilson <anne at unidata.ucar.edu> writes:
| 
| > But, I'm not convinced of the benefit of having a huge history file.
| > For one thing, innd does seem much faster with my new, relatively tiny
| > history file.  (Although it's grown by almost a factor of 8 in about 24
| > hours.)  And, speed is a significant concern for us.
| 
| Yes, the history file is one of the speed bottlenecks in INN, and the
| smaller you can keep it, the faster things will run.

True, but setting the cache size up can help, and having enough memory
tends to keep most of history in RAM. For a full feed certainly 2GB of
RAM is desirable, and a fairly large cache.

I keep making cache larger until the cache misses get down where I want
them. My numbering master gets only fresh articles, and has an average
history lookup of 44us, max 1.4ms. My reader servers which get requests
for a lot of old stuff have an average of 2.9ms. Since the backing
servers only see requests for stuff >2 days old, I can live with that;
making cache larger doesn't help the time, although the cache miss
percentage goes down (why?).
| 
| > My 'remember' value was (and still is) 10, and things have worked great
| > for months.  I'm happy with remembering articles for 10 days generally -
| > it's not that I want to remember them longer than that.
| 
| 10 days is a little long these days if you have a high volume, I think.

Agreed.
-- 
bill davidsen <davidsen at tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.


More information about the inn-workers mailing list