Subj : Re: RCU+SMR   Not!
To   : comp.programming.threads
From : Joe Seigh
Date : Thu Sep 08 2005 08:37 am

Alexander Terekhov wrote:
> Joe Seigh wrote:
> 
>>I just found out that there's a patent application in on SMR hazard
>>pointers so they're not really free to use.  At least as far as
>>using it for FOSS.  So RCU+SMR is not anymore.
> 
> 
> Even a granted patent would not make it so. BTW, how is the proof of 
> your perpetum mobile doing? (I mean "barrier free" hazard pointers.)
> 

Looks ok to me.  I haven't received feedback from the people who were
supposedly looking at it.  I was planning to wait for 4 way or 8 way
processors to become dirt cheap so I could demonstrate scalability and
robustness.  But it's a moot point now.

I think the way to go on lock-free is with Java and C# which has a
documented memory model and proper atomicity guarantees.  The whole
issue of how to implement a truly concurrent and efficient GC becomes
"somebody else's problem".  See how easy that was to fix?  :)

-- 
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.

.