TDB historical performance (with pretty graphs)

[Rusty Russell]

On Fri, 19 Jun 2009 11:20:54 am tridge at samba.org wrote:
>  > we seem to suck recently on the "replace 5000 4byte key 4 byte date
>  > records with 8 byte dataa records"
>
> That is indeed a puzzle. If I'm reading it right, then we slowed down
> by 4x between git 77dab7 and 8c8820 for the 4 byte to 8 byte expand.
>
> I can't see anything in the code history that could explain
> that. Maybe you could re-check that one manually?

Yep, I re-measured all the later commits.  There's similar weirdness in the 
results; they're not consistent.

Manually remeasuring seems to reveal that the change which "don't merge right" 
commit caused the regression in the expand test.  Presumably caused by 
freelist growth.

Rusty.