a 100% lock-free PDR algorithm for Boost, no patent issues...
Here are the links: http://groups.google.com/group/comp.programming.threads/msg/5d8201e86eb64d2a http://groups.google.com/group/comp.arch/browse_frm/thread/b2f751d5dad6c07b Here is the prototype code in C and IA-32 MSVC inline assembly language: http://home.comcast.net/~vzoom/demos/pc_sample.c IMHO, this would be a major asset to the Boost arsenal! Please, search c.p.t for the terms 'PDR' and 'proxy collector' Search, read, and read some more... Then, tell me what ya' yall think! ? ;^)
"Chris Thomasson" <cristom@comcast.net> writes:
IMHO, this would be a major asset to the Boost arsenal!
Sounds interesting, if it works. I'm not expert enough to do a deep analysis.
Please, search c.p.t for the terms 'PDR' and 'proxy collector'
Search, read, and read some more...
Well, I did a bit of that. Realistically speaking though, in this forum, you need to make some claims in lay terms to make it worthwhile for people to do the searching. Like "I have an implementatation of an algorithm that makes it easy to write ridiculously efficient lock-free datastructures." or something. A little more introductory text that explains why we should care would go a long way. -- Dave Abrahams Boost Consulting www.boost-consulting.com
"David Abrahams" <dave@boost-consulting.com> wrote in message news:873b8fa5wz.fsf@pereiro.luannocracy.com...
"Chris Thomasson" <cristom@comcast.net> writes: [...]
Yeah. It works. Its not the ideal solution, it makes use of atomic operations and memory barriers. There are other techniques that can get around those expensive instructions. However, the exotic Reader / Writer algorithms problem have patents. I have a commercial library. Joe Seighs RCU-SMR has IBM patent applications tied to them. This solution will be good for Boost, because its one of the only other alternatives. You have to hunt your way through the patents. Do a patent search on lock-free...
Well, here is some brief introductory text that describes some things: http://appcore.home.comcast.net/vzoom/ http://appcore.home.comcast.net/vzoom/round-1.pdf http://appcore.home.comcast.net/vzoom/round-2.pdf The PDR algorithm for Boost can be lock-free, but I don't think it can be Virtually Zero-Overhead... Anyway, I think the lock-free algorithm would be good for Boost. I am planning to release some test applications that make direct use of the working proxy collector prototype I linked to. Look out for them. I also have to finish that atomic static initialization document I am working on, and off: http://appcore.home.comcast.net/vzdoc/atomic/static-init/ ;^) Any thoughts?
participants (2)
-
Chris Thomasson -
David Abrahams