Initial benchmarks for checkout performance improvements

[Andrew Bennetts]

Aaron Bentley wrote:
> Aaron Bentley wrote:
> > In this run of benchmarks, cp was actually slower than copytree or
> > top-rename.  But cp has previously hit 2.183s.
> 
> The variability of results has been bothering me.  debboy in #bzr
> reminded me about the sync command last night.  I decided to try syncing
> before each run.  What a difference!
> 
> Clearly, the dirty buffers from previous runs were interfering with
> subsequent runs, because the numbers are now much lower, and more
> consistent.

That makes sense.

On a related thought, if you want to measure performance with a cold cache,
there's "echo 1 | sudo tee -a /proc/sys/vm/drop_caches".

[...]
> I'm inclined to exclude the sync time from the numbers, at least for
> day-to-day benchmarking.  I'm guessing that's more reflective of user
> experience than the other way around.

That sounds reasonable to me.  Users don't habitually type "sync" or otherwise
wait for disk IO that the OS has decided it can postpone transparently.

-Andrew.