Subj : Re: performance of single mutex versus Read/write mutex: any numbers? discuss
To   : comp.programming.threads
From : Chris Thomasson
Date : Mon Oct 10 2005 04:59 pm

> It contains demo test code which attempts to justify the use case for
> read/write and even read/write/upgradable locks.  Timing results for a
> dual processor machine are presented.

You should really take Joes advice and measure how many iterations/reads a 
thread can perform per-second. That is basically the only way to get some 
"real" performance numbers...

:)




> The link also describes a "lock
> reduced" mutex which avoids locking an OS-based mutex if there is no
> contention, although there is no code, just hand-waving.

Here is some experiential pseudo-code for a fast-pathed rw-mutex:

http://groups.google.com/group/comp.programming.threads/msg/816a18f5ea5c3bda?hl=en

This seems to outperform nearly all non-distributed rw-mutex implementations 
I have tested against; unfortunately, most of them use mutexs and condition 
variables to manage their internal bookkeeping. They "all", including the 
fast-pathed ones, seem to tank once you put them in a "contentious 
situation". IMO, the atomic operations and StoreLoad memory barriers make it 
difficult for rw-mutex's to scale...

.