Subj : Re: Best locking methods
To   : comp.programming.threads
From : Chris Thomasson
Date : Thu Apr 07 2005 01:42 am

> IMHO COW and RCU would be slow with this data structure
> since they seem to use a temporary copy, and I think the overhead
> would be too costly.

You do not have to use a temp copy to access data in RCU. You simple enter a 
RCU garbage collected region and load a pointer to the shared data-structure 
with load + data-dependant hint memory barrier. Then you can use it "within" 
the garbage collect region.

http://groups-beta.google.com/group/comp.programming.threads/msg/cdde8dea7b90ca4f




> Mutex is too slow; would need something faster than Critical Section.
> I've yet to study how SMR and atomic_ptr do work.

SMR has algorithm that requires expensive memory barriers. However, you can 
layer atomic pointers on top of an SMR implementation. You would pay for 
accessing the SMR API when you copy a shared pointer to a local pointer.

http://groups-beta.google.com/group/comp.programming.threads/msg/cdde8dea7b90ca4f




Also, SMR is fragile. You really need to implement in assembly language...

http://groups-beta.google.com/group/comp.programming.threads/msg/423df394a0370fa6


-- 
http://appcore.home.comcast.net/
(portable lock-free data-structures)

.