[GHC] #7606: Stride scheduling for Haskell threads with priorities

[GHC]

#7606: Stride scheduling for Haskell threads with priorities
---------------------------------+------------------------------------------
    Reporter:  ezyang            |       Owner:  ezyang          
        Type:  feature request   |      Status:  new             
    Priority:  normal            |   Milestone:  7.8.1           
   Component:  Runtime System    |     Version:  7.7             
    Keywords:                    |          Os:  Unknown/Multiple
Architecture:  Unknown/Multiple  |     Failure:  None/Unknown    
  Difficulty:  Unknown           |    Testcase:                  
   Blockedby:                    |    Blocking:                  
     Related:                    |  
---------------------------------+------------------------------------------

Comment(by ezyang):

 Replying to [comment:2 simonpj]:
 >  * What the goal of this change is.

 The goal of the change is to allow Haskell programmers to change how
 frequently threads they create are being run; in case there are some
 threads that are important and should be run as frequently as possible, or
 threads are being created by an untrusted user who should not be allowed
 to thread-bomb the system.

 >  * Roughly how the new scheduling algorithm achieves the goal.

 Stride scheduling operates by replacing our old doubly linked list run
 queue with a priority queue. The priorities are 'passes' which increment
 by their stride (computed based on their priority) every time they run;
 thus, a task with a low stride will run multiple times before their pass
 catches up.

 >  * There seems to be something about priorities involved.  Who sets
 those priorities? The programmer?  The runtime system?

 The priorities are set by the programmer using setTickets and related
 functions.

 >  * What about priority inversion?

 Without any change to priorities, scheduling behavior is intended to be
 identical to what it was previously.  Priority inversion can occur when a
 thread blocks on another; this patch doesn't address that yet (though I've
 mentioned about how to do it for BLACKHOLES).

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7606#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler