Interesting. So, if my Mac Carbon VM on 10.3 is still only yielding ~50 requests per second with Seaside, is that still likely a networking issue (PowerBook G4 867MHz)?

On Dec 7, 2003, at 7:15 PM, John M McIntosh wrote:

Andreas found an issue with the windows code. When the windows VM would sleep it would not awake for incoming network events, that was the issue Andreas referred to. No changes to the macintosh or linux VM for this

On Dec 7, 2003, at 3:53 PM, Phil Hargett wrote:

...

Great, John! I'll update my sources in the next few days, and see how it works.

Question for you: does this edition include any tweaks to enhance network performance? I recall last month there was a discussion about Seaside performance (as measured by the Apache benchmark tool) that appeared to suggest the handling of semaphores for sockets was not optimal. Any enhancements included in this build of the VM?

Thanks again! :)

On Dec 7, 2003, at 5:33 PM, John M McIntosh wrote:

...

I've pushed a Mac VM 3.7.0b2 for testing to

http://homepage.mac.com/johnmci/

Squeak 3.7.0Beta2.app.sit

It contains: a) You can alter the application's Info.plist by setting the SqueakEncodingType using the ICANA readable names

macintosh - >macRoman UTF-8 -> UTF-8

b) Support for sqSocketListenOnPortBacklogSizeInterface, which allows you to specify the interface to use for the listen should work both for os-x and os-9. c) Usage of VMMaker3-7a2. d) Special change sets: make primitive point x/y internal, and slightly faster activate.

--

====== John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===================================================================== ======

--

==== John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================= ====

The current mac vm's use the unix network code so they should exhibit the same behavior as the the unix VM. You could try the classic mac VM, by showing the applications contents and double clicking on the Squeak 3.7.0b2 Classic found in the MacOSClassic subfolder. That still uses Open Transport, which is I suspected layered on the BSD Socket code. Still it might be interesting to see if they have the same thruput numbers.

On Dec 7, 2003, at 4:58 PM, Avi Bryant wrote:

On Dec 7, 2003, at 4:51 PM, Phil Hargett wrote:

...

Interesting. So, if my Mac Carbon VM on 10.3 is still only yielding ~50 requests per second with Seaside, is that still likely a networking issue (PowerBook G4 867MHz)?

FWIW, I got the best results by:

• using the unix VM
• killing the UI process (Project uiProcess suspend)

-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===

Mmm I'll bet it's tied to aioPoll() calling frequency.

In the carbon VM, this is called either at event processing fetch time, or at idle time. When you turn off the uiProcess and have a busy image that doesn't get idle time, then then no calls to aioPoll(). However the EventSensor process does make a call to get next event every 500 ms giving you I suspect your 2 rps per second you see when you've suspended uiProcess

If you set Preferences higherPerformance on, do you get a better rate?

If this is true we'll need to think about a better way to call aioPoll()

On Dec 7, 2003, at 9:31 PM, Avi Bryant wrote:

On Dec 7, 2003, at 5:06 PM, John M McIntosh wrote:

...

The current mac vm's use the unix network code so they should exhibit the same behavior as the the unix VM.

I'm comparing 3.7.0Beta2 (Carbon) to 3.6g-2 (Unix). With the UI process running, they both get similar results (the particular test I was doing got about 28 requests per second with the Carbon VM, 32 rps with the Unix VM, but that difference may not be meaningful).

If I suspend the UI process, I get about 45 rps with the Unix VM, but the Carbon VM drops to 2.00 rps.

So it may not be the network code, exactly, but there's something wonky in the event handling.

Avi

-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===