[concurrency-interest] David Ungar talk: Everything You Know about Parallel Programming Is Wrong!
ron.pressler at gmail.com
Fri Mar 9 07:09:22 EST 2012
Just wanted to let you know about this video of a talk by David Ungar from
He basically discusses a new concurrent programming style for many-core
systems that completely does away with concurrency primitives, including
CAS. The idea is to do everything you can to mitigate the chance of a race
condition, and then, if a race happens, try to fix it later and/or live
with some wrong results.
What I found particularly interesting is the "probabilistic linked-list" he
discusses towards the end, where he shows a simple data structure that
might miss a few inserts, but on the whole has many more "successes per
millisecond" (a metric he thinks is valuable) than CAS-ing. He doesn't
mention memory fences at all, though, which I found a little strange.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Concurrency-interest