> Here is some pretty good info on lock-free structures... I'm pretty
> sure I tested their code in a multithreaded high-concurrency
> environment and experienced the problems I was discussing.
I understand.
The algorithm is quite complex.
The old version was not really fast.
In the paper cited, some tests was done with 90 concurrent threads, and
with different density of data inside the structure. I found really
interresting (but I am not expert of this field).
Most of the researcher that work on "concurrent lock free" algorithm
understand the complexity, and to help anyone to use it, they provide
this technology for a whole algorithm.
There are libraries for red-black tree, skip-list, hash-table, etc.
I did not read the detail of the code to know if we may use it as a
direct remplacement.
http://www.cl.cam.ac.uk/Research/SRG/netos/lock-free/
http://www.cl.cam.ac.uk/users/kaf24/lockfree.html
http://www.audiomulch.com/~rossb/code/lockfree/
http://www.cs.rochester.edu/u/scott/synchronization/
Cordialement,
Jean-Gérard Pailloncy