[threadpool] work-stealing support
Hello, I've the work-stealing strategy ready - I'd like to know if it is desired to remove lazy and adaptive strategy from the lib (as Vicente Botet and Johan Torp mentioned)? So only fixed an work-stealing will be supported?! thx, Oliver
----- Original Message ----- From: <k-oli@gmx.de> To: <boost@lists.boost.org> Sent: Monday, September 29, 2008 11:23 AM Subject: [boost] [threadpool] work-stealing support
Hello, I've the work-stealing strategy ready - I'd like to know if it is desired to remove lazy and adaptive strategy from the lib (as Vicente Botet and Johan Torp mentioned)? So only fixed an work-stealing will be supported?!
Hi Olivier, It is a very good new that you have incorporated work-stealing in your library. Can we get the current state somewhere? In my opinion neither the lazy nor the adaptive should be very useful, so from my side you can remove them. This will simplify your library. Thanks for incoporating work-stealing, it is a major feature for me. Vicente
Am Montag, 29. September 2008 12:32:11 schrieb vicente.botet:
Hello, I've the work-stealing strategy ready - I'd like to know if it is desired to remove lazy and adaptive strategy from the lib (as Vicente Botet and Johan Torp mentioned)? So only fixed an work-stealing will be supported?!
Hi Olivier,
It is a very good new that you have incorporated work-stealing in your library. Can we get the current state somewhere?
In my opinion neither the lazy nor the adaptive should be very useful, so from my side you can remove them. This will simplify your library.
Thanks for incoporating work-stealing, it is a major feature for me.
Vicente
Hello Vicente, thanks for your response - I'll remove lazy and adaptive strategy and release the new version in the boost vault soon. Please note that the work-stealing is not fork-join as you requested in some of your previous emails (so please don't be disapointed). For fork-join two possible strategies are possible - one like in JAVA (from Doug Lea) which blocks in pool::submit() until the main task and its sub-tasks are finished - the other solution would require stack switching as it is provided by the coroutines lib in the vault. I would prefere the stack switching solution but it seams to become a complex implementation. regards, Oliver
----- Original Message ----- From: <k-oli@gmx.de> To: <boost@lists.boost.org> Sent: Monday, September 29, 2008 1:01 PM Subject: Re: [boost] [threadpool] work-stealing support
Hello Vicente, thanks for your response - I'll remove lazy and adaptive strategy and release the new version in the boost vault soon. Please note that the work-stealing is not fork-join as you requested in some of your previous emails (so please don't be disapointed).
No problem. I 'm looking for the feature not the form.
For fork-join two possible strategies are possible - one like in JAVA (from Doug Lea) which blocks in pool::submit() until the main task and its sub-tasks are finished - the other solution would require stack switching as it is provided by the coroutines lib in the vault. I would prefere the stack switching solution but it seams to become a complex implementation.
Yes, I understand. A thread pool on top of the coroutine library is an excelent idea. Best, Vicente
participants (2)
-
k-oli＠gmx.de -
vicente.botet