Subj : Re: posix and lock-free algorithms
To   : comp.programming.threads
From : David Schwartz
Date : Thu Aug 11 2005 02:46 pm


"Alexander Terekhov" <terekhov@web.de> wrote in message 
news:42FB3F3D.C3611585@web.de...

> David Schwartz wrote:
> [...]
>>     Since you must lock and unlock a mutex every time you pass through 
>> this
>> code, it's really not DCL. It's just normal locking with extra overhead.
>> Hypothetically, there could be reduced contention.

> If with "reduced contention" you mean the use of per-thread locks, it's
> totally busted just like the DCL classic. DCSI-MBR just can't be done
> using per-thread locks to substitute msync::ssb -> msync::{dd}{cc}hlb
> protocol on atomic pointer or flag.

    Why not? POSIX defines the mutex functions as sychronizing memory. 
(Except, of course, for the issue that it makes concurrent read/write 
undefined.)

    DS

.