Weird Windows 2000 scheduling behaviour with Squeak VM
Stephen Pair
spair at advantive.com
Thu Jan 24 17:25:19 UTC 2002
I've also seen this behavior on Win2k and WinNT. It's only in more
recent VMs. Squeak 2.5 did not exhibit this behavior.
- Stephen
> -----Original Message-----
> From: squeak-dev-admin at lists.squeakfoundation.org
> [mailto:squeak-dev-admin at lists.squeakfoundation.org] On
> Behalf Of Mike Rutenberg
> Sent: Thursday, January 24, 2002 1:00 PM
> To: Squeak-dev at lists.squeakfoundation.org
> Cc: Andreas.Raab at gmx.de; Mike Rutenberg
> Subject: Weird Windows 2000 scheduling behaviour with Squeak VM
>
>
> If I do a long Squeak calculation like
> [true] whileTrue: [12 / 4]
> the Windows 2000 scheduler will only occasionally give any
> other "normal priority" Win32 process any CPU time. Other
> processes, even those who are "on top", will be almost
> entirely deprived of cycles. For example, the TaskManager
> CPU graph or list of running processes no longer updates
> regularly, often pausing for 20 seconds. Similarly
> InternetExplorer and the FileManager become very unresponsive.
>
> I have also noticed the same behavior if I compact a Celeste
> mail database (which involves lots of memory allocations and
> IO) rather than an endless simple loop.
>
> Interestingly, a second simultaneously running Squeak VM is
> not affected while the first runs the endless loop. It
> responds normally though all of this, and does not seem
> noticeably slowed down, for instance quickly opening a change sorter.
>
> When I am able to finally bring up the TaskManager, I can
> eventually lower the priority of Squeak.exe and the system
> immediately becomes responsive again.
>
> As a test, I wrote a small C program to duplicate this endless loop.
> When I run it, the CPU meter pegs at 100%, but the system
> does not become unresponsive.
>
> Has anyone else seen anything like this under Windows NT or 2000? Is
> this a side effect of something done intentionally in the VM?
> I do not
> remember seeing problem under Windows 98.
>
> Mike
>
More information about the Squeak-dev
mailing list
|