[boost] [lockfree] Faster ring-buffer queue implementation

Subject: [boost] [lockfree] Faster ring-buffer queue implementation
From: Alexander Krizhanovsky (ak_at_[hidden])
Date: 2014-01-15 13:32:19

Hello,

some time ago I've implemented lock-free multi-producer multi-consumer 
queue on ring buffer and recently I've compared it's performance with 
current (Boost 1.55.0) boost::lockfree::queue - it shows 50% better 
performance. I've made just a quick glance on the Boost implementation 
and noticed that it uses CAS operations which is basically slower than 
simple RMW (atomic increment in our case) on modern x86-64 hardware.

Maybe I've made some mistake in boost::lockfree::queue usage. If so then 
I'll be obliged for pointing me out the problem in my benchmark. 
Otherwise I'm wondering if it has sense to integrate the queue 
implementation into boost::lockfree?

The algorithm description is available at 
http://www.linuxjournal.com/content/lock-free-multi-producer-multi-consumer-queue-ring-buffer 

The source code of the queue with the benchmark for 
boost::lockfree::queue: 
https://github.com/krizhanovsky/NatSys-Lab/blob/master/lockfree_rb_q.cc
And the benchmark results are available here: 
http://natsys-lab.blogspot.ru/2014/01/natsys-lock-free-queue-in-comparison.html 

Thanks!

-- 
Alexander Krizhanovsky
NatSys Lab. (http://natsys-lab.com)
tel: [phone]
e-mail: ak_at_[hidden]