Re: [Chicken-users] newbie thread questions?

[F. Wittenberger]

Am Donnerstag, den 11.12.2008, 09:26 +0100 schrieb Basile STARYNKEVITCH:
> Hello All
> 
> 
> (my second mail of questions; I'm very newbie!)
> 
> Apparently the Chicken thread package is continuation based and does not 
> use pthread hence is not able to take of profit of several cores running 
> several pthreads in the same unix process.

Yes.

> Are there any Chicken packages which are blocking? More specifically, 
> can I run, using the http package, both HTTP client & HTTP server & some 
> other user code without being blocked by a slow system call?

Don't know.  Those I did not use so far.

> Can I suppose that every chicken thread syncrhonisation primitives (in 
> particular mutex) are very cheap? My intuition is that they are 
> user-level function without any system calls (ie to futex) inside!

You can.

That is, before you do, you want to include this in your code and
compile it with (disable-interrupts) and maybe some more declarations
you can lift from srfi-18.scm:

(global-set! 'mutex-unlock!
  (lambda (mutex . cvar-and-to)
    (##sys#check-structure mutex 'mutex 'mutex-unlock!)
    (let ([ct ##sys#current-thread]
          [cvar (and (pair? cvar-and-to) (car cvar-and-to))]
          [timeout (and (fx> (length cvar-and-to) 1) (cadr cvar-and-to))] )
      (dbg ct ": unlocking " mutex)
      (when cvar (##sys#check-structure cvar 'condition-variable
'mutex-unlock!))
      (##sys#call-with-current-continuation
       (lambda (return)
         (let ([waiting (##sys#slot mutex 3)]
               [limit (and timeout (##sys#compute-time-limit timeout))] )
           (##sys#setislot mutex 4 #f)
           (##sys#setislot mutex 5 #f)
           (##sys#setslot ct 8 (##sys#delq mutex (##sys#slot ct 8)))
           (when cvar
                 (##sys#setslot cvar 2 (##sys#append (##sys#slot cvar 2) 
(##sys#list
ct)))
                 (cond [limit
                        (##sys#setslot 
                         ct 1
                         (lambda () 
                           (##sys#setslot cvar 2 (##sys#delq ct (##sys#slot 
cvar 2)))
                           (if (##sys#slot ct 13) ; unblocked by timeout
                               (return #f)
                               (begin
                                 (##sys#remove-from-timeout-list ct)
                                 (return #t))) ) )
                        (##sys#thread-block-for-timeout! ct limit) ]
                       [else
                        (##sys#setslot ct 1 (lambda () (return #t)))
                        (##sys#setslot ct 3 'sleeping)] ) )
           (unless (null? waiting)
                   (let* ([wt (##sys#slot waiting 0)]
                          [wts (##sys#slot wt 3)] )
                     (##sys#setslot mutex 3 (##sys#slot waiting 1))
                     (##sys#setislot mutex 5 #t)
                     (when (or (eq? wts 'blocked) (eq? wts 'sleeping))
                           (##sys#setslot mutex 2 wt)
                           (##sys#setslot wt 8 (cons mutex (##sys#slot wt 8)))
                           (when (eq? wts 'sleeping) (##sys#add-to-ready-queue 
wt) ) ) ) )
           (if (eq? (##sys#slot ct 3) 'running)
               (return #t)
               (##sys#schedule)) ) ) ) ) ))


Moreover please consider to try the attached scheduler.scm as a drop in
replacement for the current one.  Especially when your are using
multithreaded code with a lot of i/o.  Heavy HTTP/S communication are
the context where I found the shortcomings of the current scheduler.
YMMV however.  When you did, please drop me a note about the effect in
your case.

Best regards

/Jörg

scheduler.scm
Description: Text Data