For interest: Bugzilla performance

[Gervase Markham]

Myk Melez wrote:
> Gervase Markham wrote:
> 
>> Pre-changes:  45.9 mails per day over 7 days
>> Post-changes: 16.3 mails per day over 12 days 
> 
> Note that the number of killed queries varies widely by day of week 
> (f.e. there were 6 on Sunday, 17 on Monday, and 50 on Tuesday), so 
> measuring two different time periods is likely to throw off the data.

I hoped to counteract this problem by averaging. The second twelve days 
followed on from the first seven, so (roughly) we have a single week 
average, then a following two-week average.

> Note also that emails sometimes mention multiple killed queries, 
> especially when the queries are generated by a DOS attack.

This is also true, and I made no attempt to count this. I assumed, 
perhaps incorrectly, that this would not affect one set of results any 
more than another.

> We stopped a DOS attack last Wednesday evening, not the first we've 
> dealt with in the last month, and that may have skewn results both 
> before and afterwards.

If you exclude last Wednesday from the results (65 mails), then that 
16.3 figure goes down to 11.9, making the improvement greater.

>> Further performance changes, which required changes to Bugzilla's 
>> code, will kick in at the upgrade on November 8th.
> 
> Unfortunately we'll inherit some performance regressions as well from 
> some recent code changes (the groups rewrite in particular) that the 
> performance improvements will hopefully counteract.

Indeed. It's all swings and roundabouts.

Gerv