Lock-free tdb lookups

Joe Seigh jseigh_02 at xemaps.com
Tue Jan 17 14:56:58 GMT 2006


tridge at samba.org wrote:
> Joe,
> 
>  > Whatever happened there?  Was or is tdb performance an issue or
>  > was this just considered an experiment?  Just curious.
> 
> I can't find any earlier message on this - did I just miss it? I'd
> certainly be interested in any workable lock-free tdb proposals. My
> recent tdb changes reduced the number of locks needed for some common
> operations by quite a lot, but it definately isn't lock free.
> 
There was a dicussion with either you or Jeremy, Paul McKenney, Rusty,
and myself back around April 2004 about using RCU for preemptive user
threads for the tdb but it fell through since Rusty wasn't available
to do the work in the Linux kernel.

I was wondering whether the tdb was a valid example of something where
lock-free would help or whether it was just something Paul wanted to
try out.  I have a couple of other ways to do lock-free reads here
http://atomic-ptr-plus.sourceforge.net/ notably the RCU+SMR (fastsmr)
and reference counted proxy collector (rcpc) stuff which are in theory
portable to about anything (although I only have 32 bit systems at the
moment).  The problem in working with scalability is it usually involves
large scale things which I can't duplicate at home very easily, so I'm
trying to figure out whether whether the problem exists in the wild in
a form that doesn't require major infrastructure to implement or having
Comcast get on my case for abusing their network.

--
Joe Seigh



More information about the samba-technical mailing list