Subj : Re: performance of single mutex versus Read/write mutex: any numbers? discuss
To   : comp.programming.threads
From : Howard Hinnant
Date : Mon Oct 10 2005 09:41 pm

In article <T_KdnT4EgPD7A9neRVn-uw@comcast.com>,
 "Mike Winter" <mike.winter@comcast.net> wrote:

> If I have a small critical section to access an element of a structure 
> (queue) and most accesses are reads and once in a blue moon a write (PI 
> seconds is a nano-century!) occurs where the primary element is superceded, 
> but only completely superceded when its reference count is 0(no more 
> readers), I think a single mutex may be overkill. I would think that a 
> single mutex would be less performant than say a more complex structure like 
> say a read-write mutex. Then I think about what it takes to manage a 
> ReadWriteMutex (mutex + reader-count + writing).
> 
> Consider
> 
> struct ReadWriteMutex {
>       Mutex    mWriterMutex;    // gets grabbed rarely
>       AtomicCounter mReaderCount;  // gets incremented/decremented securely 
> and fast
>       ReaderMutex  mReaderMutex;   // only blocks if mWriterMutex is locked
> };
> 
> Thats the context/concept.
> 
> Heres the question: Are there any numbers /papers in the ether that can 
> point to how the ReaderWriterMutex will compare to the ordinary Mutex? Links 
> welcomed
> 
> MW

This link might be helpful:

http://home.twcny.rr.com/hinnant/cpp_extensions/rw_perf.html

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.  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.

At best this paper should only convince you that perhaps it is worth the 
trouble to code it up both ways and measure.

-Howard

.