Re: [Monotone-devel] nvm.sqlite3.binary branch created

[Nathaniel Smith]

On Wed, Dec 14, 2005 at 04:21:31PM +0100, Christof Petig wrote:
> while the other developers are busy exploring rosters I decided to reach
> for some low hanging fruits: optimizing the database size
> 
> Since I have lot of databases a 20-30% reduction in size would make a
> difference to me. Whithin one work hour I was able to show that binary
> content is possible and simply unbase64'ing file(_delta)s show a 15%
> size reduction on the database.

Great!

You may want to work off of the rosters branch, rather than mainline;
there are a number of changes in the rosters branch to the database
storage (in particular, the db doesn't store manifests anymore, and it
does store rosters!).  That should reduce the amount of work you have
to do after rosters land.

> PPS: I will not touch the IDs (hexencoded sha1) _first_ but having
> database indices and access with _halved_ keylength should give a
> performance gain. I will _measure_ and report after tackling the base64
> encoded BLOBs.

We've also discussed some the idea of switching to a kind of
"interned" representation for revision ids, where most places that use
them just store a uint32, and there's a separate table to get the real
id for the times when you need to print it or whatever.

> P3S: schema_migration.cc looks awfully verbose.

What do you mean?

-- Nathaniel

-- 
IBM manual SENG-5155-01:  Power Supply and Air Moving Device Installation
Instruction for iSeries 820 and 5075.