Subj : Re: Best locking methods
To   : comp.programming.threads
From : Sean Burke
Date : Thu Apr 07 2005 07:34 pm


"Joe Seigh" <jseigh_01@xemaps.com> writes:

> On Thu, 7 Apr 2005 04:56:26 +0200, Uenal Mutlu <520001085531-0001@t-online.de> wrote:
> 
> >
> > Say, it is a sorted vector of structures. The vector has to grow
> > when new items need to be inserted.
> (...)
> >
> > Each item is a fixlen structure consisting of a unique key, plus its data.
> > Obviosly the writer threads are those who update the list.
> > The readers are free how they access the list; besides using a Find(key)
> > function they can also walk sequentially in the list, ie. in a for loop.
> > Access to an item is done via baseptr + the physical element number ("record
> > number").
> >
> > This works fine using Critical Section's, but since there seem to be many
> > methods available I wondered whether there's maybe a significantly faster
> > method.  What would others use for such a task?
> >
> >
> As to whether there's a faster method for mutual exclusion than Critical
> Section, don't you think they would be using it instead if it existed?
> 
> Also, you've picked a data structure, vector, that's rather expensive to
> modify.  A binary tree with PCOW (Partial Copy On Write) would work
> better but it's a rather advanced project.  I don't think anyone has
> done one yet and it will probably be patented so you won't be able to
> use it for free and it will be too late to do a free implementation then.

Unless it is important to be able to walk keys in sorted order,
it seems like a hash table with per-bucket locking would offer
both better performance and improved concurrency.

-SEan

.