ldb speed

tridge at samba.org tridge at samba.org
Mon Oct 30 22:01:04 GMT 2006


 > That was the idea, even easier consider dragging a couple of groups
 > containing 10000 users ... (means changing the member attribute on all
 > of them).

All of these 'what if' scenarious miss one very obvious thing. None of
them will cost more than totally rewriting the database on disk in
order to complete the operation. That is our 'worst case' limit, and
that limit is not measured in days or hours, its measured in seconds. 

I think it will be _extremely_ rare that we actually have to totally
rewrite the database due to a user operation, but it is a firm upper
limit on complexity, and that upper limit is well within the range
that would be acceptable.

You do know that right now every time you change the indexing of a
ldb+tdb it rewrites the entire database? It could do it in a smarter
way, but we don't because even with quite a large ldb that rewrite
takes a fraction of a second. You don't even notice it in most cases! 

If we try to make stuff 'smart' underneath to cope with these rare
operations in a supposedly efficient manner then what will happen is
that the operations that we really care about will get slower. It is
not worth it.

Cheers, Tridge

More information about the samba-technical mailing list