Re: [Monotone-devel] Re: times to load various things from the database

[Markus Wanner]

Hi,

Derek Scherger wrote:
> The following patch improves this by a factor of 3:
> 
> Sat Jan 17 09:47 PM ~/monotone/mainline $ time ./mtn db load_rosters
> mtn: loading rosters
> mtn:       rosters
> mtn: 14,541/14,541
> 
> real    7m55.062s
> user    7m46.029s
> sys     0m9.061s

Sounds good.

I've been thinking about load_rosters in general. It is currently
benchmarking the exact use case of git_export: loading all rosters in
forward direction. But that's not exactly what we should optimize for,
is it?

So, what's the common use case for standard mtn operations? Which ones
do need rosters at all? What rosters do they need? Can (and should) we
extend the load_rosters command to simulate other access patterns (like
i.e. random loading)?

> I'm not sure if this is likely to cause problems with other aspects of
> performance or not. I'll try timing the testsuite, a full pull and maybe a
> db check as well with and without to see what it does.

Hm.. extending nvm.cbench is getting back on our TODO list, I guess.

Regards

Markus Wanner