[3923] in java-interest
Re: Thread Scheduling in Win32
daemon@ATHENA.MIT.EDU (David Strauss)
Sun Dec 3 14:37:26 1995
Date: Sun, 3 Dec 1995 12:08:26 -0500 (EST)
To: java-interest@java.sun.com
From: dstrauss@panix.com (David Strauss)
In article <49hgm2$625@romeo.logica.co.uk>, stupplem@logica.com says...
>
>In article <49hf85$cmi@news1.panix.com>, dstrauss@panix.com says...
>>
>>There's an applet called RaceTest that accompanies the section on
>>thread priorities in the Java Progammer's Guide. It creates two
>>"Runner" threads whose run method is a loop that counts up to
>>400K. The runners do no I/O so they should not block. There is a
>>higher priority thread that occasionally wakes up and reports the runers'
>>progress.
>>
>>According to both the Java and Win32 documentation, thread scheduling
>>is done only among those threads that are at the highest
>>priority level of all currently runnable threads. Thus, if you
>>set the runners to different priorities, the lower priority
>>runner should be "starved out" until the higher one finishes.
>>This is clearly not happening when I run the applet in Windows 95.
>>Although the higher priority runner always wins by a large margin,
>>the lower priority thread manages to butt in every once in awhile and
>>bump its count a few ticks.
>>
>>This behavior is contrary to both the Java and the Win32 specs. Does
>>anybody know what's going on? (I did discover that the 10 Java
>>priorities are mapped to only 6 Win32 priorities and I am
>>definitely assigning different Win32 priorities to the runners,
>>so it has nothing to do with that.)
>
>Not sure about java scheduling, but as far as Win32 stuff goes I think you
can
>only ever set the base priority level of a thread - the dynamic priority
>(which is actually the one used for scheduling) is assigned by the system
>based upon several factors, including things like total CPU use and time
since
>last timeslice ...? ... or something like that.
>This ould explain the bahaviour you see - the higher priority racer gets most
>CPU, but as it does so it's dynamic priority is reduced and the dynamic
>priority of the 'tortoise' thread is increased allowing it to occasionally
get
>a timeslice.
>
>Anyone else care to add to - or refute - any of the above?
>
>Matt.
>--
>Matt Stupple - stupplem@logica.com - http://www.logica.com/
>Logica UK Ltd, 75 Hampstead Rd NW1 2PL, +44(0)1716379111
>Words, thoughts and ideas all my own.
>
This certainly makes sense out of the behavior I'm seeing. If this
is true then there is no way in Win32 to "starve" a lower priority
process?? It seems that there should be.
--
------------------------------------------------
David Strauss
E-mail: dstrauss@.panix.com
dstrauss@lehman.com
-------------------------------------------------
-
This message was sent to the java-interest mailing list
Info: send 'help' to java-interest-request@java.sun.com
