Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming.threads > #2790

About task-based scheduler

Show all headers | View raw

Hello,


I have just read the following webpage by Dmitry Vyukov:

http://www.1024cores.net/home/parallel-computing/radix-sort

Please read the "Scheduling" part, you will notice that Dmitry Vyukov
have implemented a task-based scheduler on top of the Win32 threading 
API. In main part it's similar to classical Cilk-style work-stealing 
scheduler that supports system-topology awareness, hyper-threading 
awareness, affinity-awareness, batch-spawn capability and manual 
task-depth control, but let us ask an important question: Why
Dmitry Vyukov have invented this task based scheduler ? of course
it is to optimize more , but the question that i am asking is
that i think that this task-based scheduler that look like the Cilk 
scheduler is especially designed for "recursive" calls in parallel
programs, so if you take a look at my new parallel sort algorithm,
its parallel sorting part is not using recursive calls and its parallel 
merging part is not using recursive calls, so i don't think  Dmitry's 
Scheduler will bring improvement to my parallel algorithm, other
than that even if i rewrite my parallel algorithm to use recursive calls
this will not bring improvement to "big" data and large industrial data 
and such, cause what will do  the Dmitry's scheduler is to optimize for 
a small part of the overall big data, so the Dmitry's scheduler is not 
suitable for general cases, so i don't think i will implement such 
scheduler , so i will stay with my current threadpool that i have 
implemented and used in my parallel algorithms.



Thank you for your time.



Amine Moulay Ramdane.

Back to comp.programming.threads | Previous | Next | Find similar | Unroll thread

Thread

About  task-based scheduler Ramine <ramine@1.1> - 2015-01-02 20:54 -0800

csiph-web