[Monotone-devel] Performance of the OpenEmbedded database (was Re: Do we really need to IDN-encode cert names, keypair ids, and database variable domains?)

[Ludovic Brenta]

Koen Kooi writes:
>> I'll do that in the future, but the snapshot is not that much bigger
>> than my local monotone database:
>> 
>> -rw-r--r-- 1 zack zack 154M 2007-10-20 11:23 monotone.mtn
>> -rw-r--r-- 1 zack zack 183M 2007-10-20 21:39 OE.mtn
>
> Try doing operations on the org.openembedded.dev branch (e.g. a clean
> pull) and you'll notice it is *way* slower than pulling any net.venge
> branch :(

I too noticed that operations like pull are much slower on the
OpenEmbedded than on the Monotone or Pidgin databases, which are of
similar sizes, and I wondered why.  One possible explanation is that
the OpenEmbedded database contains much fewer rosters and more files
than the other databases (OE: 9 full rosters and 14.5E3 files; Pidgin:
22 full rosters and 6.5E3 files; Monotone: 139 full rosters and 7.3E3
files).

I speculate that the slowness is either due to the number of files
(i.e. monotone does not scale well with the number of files), or
because rosters in OE tend to be very large (i.e. monotone does not
scale well with the number of files per roster), or both.

-- 
Ludovic Brenta.