Subject: Re: [boost] Boost.Thread project for GSoC 2014
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2014-03-07 13:22:26
Le 07/03/14 08:43, Tim Blechmann a écrit :
>>>> lock-free data structures?
>>> No. But I can learn <atomic> quickly.
>>> Is lock-free data structures going to be used a lot for this project?
>> The work-stealing thread pool would have less contention if we use a
>> lock-free dequeue. Of course, we can use Boost.LockFree if it provides
>> already whatever we need.
> actually, there are work-stealing queue data structures, with private
> 'push'/'pop' and a public 'steal' function. i do have a prototype of a
> textbook-implementation from herlihy/shavit, which is pretty efficient,
> although it is bounded (the maximum number of elements has to be defined
> in advance) and optimistic (under certain conditions a 'steal' operation
> may fail).
Thanks Tim for the info.
I have not yet analyzed how a thread_pool could work with bounded queues.
Do we want to block when the queue is full?
> it would make sense to integrate such a data structure into
> boost.lockfree as part of a GSOC regarding a work-stealing thread-pool.
>
>
If we found a consensus on what to do when the queue is full, yes this
can be done during the GSoC project.
Vicente