Lock-free tdb lookups

[Joe Seigh]

tridge at samba.org wrote:
> Joe,
> 
>  > I can't see any way of making things lock-free without infringing,
>  > in part at least, on patent application 20050198030 "Efficient
>  > support of consistent cyclic search with read-copy-update" by Paul
>  > McKenney, file in March 2004,
> 
> I'd be pretty surprised if Paul's employer (IBM) wasn't happy to allow
> that patent to be used in tdb. In Paul's talk at LCA last year I think
> he specifically mentioned tdb as a possible use.
> 
> So I think the main impediment would be someone implementing it, and
> proving its safe on all the architectures we care about. We'd also
> have to decide if the extra complexity is worth it. If we find its
> only a small gain then I'd rather keep the simple approach we have
> now.
> 
> Also, with transactions there is very little locking going on anyway,
> as the operations that happen within the transaction are lockless. 
> 

A further follow up here.  I think there must have been a contention
problem at one time, otherwise you would not have gone to per chain
locking.  It's not clear whether you are still experiencing contention
and how badly.  I'm not sure it's worth expending resources to solve
a problem that may not exist.  Actually, at this point, I'm not even
sure I have the luxury of having any resources to expend on this problem
even if it did exist, interesting as it seems to be.

--
Joe Seigh