[boost] Re: Smart pointers: One more implementation +question

From: Alexander Terekhov (terekhov_at_[hidden])
Date: 2003-02-06 14:18:04

Pavel Vasiliev wrote:
[...]
> I really think that having the only mutex for all short smart
> pointer-related interlocked operations will not harm performance of
> real-life applications in mp systems. In my code this mutex is used
> only for really short operations like "lock, increment, save to
> temporary, unlock, test the temporary". It is hard to imagine that
> contention will often occur for so short operations.

Well, "raw" contention aside, with "ping-pong" I mean that with your 
design clients might rather quickly get into trouble with respect to 
"false sharing" effect(s) -- writing to "cache line" that is also 
held by another processor. The problem here is rather expensive 
communications needed to invalidate the other processor's cache-line, 
feed it with "up-to-date" data [that's most likely an additional 
"slow trip" to memory], etc. Well, you might want to take a look at: 

< "Performance Management Guide" [AIX 5L Version 5.1] >

http://publibn.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixbman/prftungd/2365c31.htm#HDRI42667
(Lock Granularity)

http://publibn.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixbman/prftungd/2365c31.htm#HDRI26126
(Cache Coherency)

http://publibn.boulder.ibm.com/doc_link/en_US/a_doc_lib/aixbman/prftungd/2365c33.htm#HDRI14109
(Multiprocessor Throughput Scalability)

< etc. "general" stuff from this publication >

regards,
alexander.