[self-interest] UI Performance Question

Michael Latta lattam at mac.com
Thu Jun 17 18:37:44 UTC 2004


Based on the comments it should have been 5, but could be 100 to reduce 
load on the CPU when idle.  So, I would not expect any bad effects by 
putting it back to 5.

There are bad effects by removing the call to this method.  The UI 
locks up.

Michael



On Jun 17, 2004, at 11:24 AM, Ben Moseley wrote:

> Wow - nice work - that seems to make a big difference.
>
> Is there any catch to this? Was it simply a mistake to set it to 100?
>
> --Ben
>
> On 17 Jun 2004, at 18:31, Michael Latta wrote:
>
>> I got the first round of timings done.  It is spending about 5% of the
>> elapsed time in the doStep method!!  The rest of the time appears to 
>> be
>> elsewhere.  Is it possible that the other processes are consuming that
>> much CPU for the timeout and watcher processes?  Or on semaphores?  Or
>> the scheduler?  Since this basically matches the CPU used by the OS
>> process it would suggest that synchronization is using the other 95%.
>> I will look at the delay in runloop to see if it is delaying when it
>> should not do so.
>>
>> I did a little more nosing around.  The problem is in
>> delayIfNoInputFor: method on traits worldMorph.  The code has been set
>> to delay in increments of 100ms!!!! So any delay results in 100ms of
>> delay at least.  By changing this to 5 (as the comments imply it 
>> should
>> be) the frame rate goes to 25 which is what you would expect with 35ms
>> frame times as requested at 5ms resolution on the delay.  The method
>> should really compute the delay when it is less than the resolution.
>> When I get the time I will post a modified version.
>>
>> Michael
>>
>>
>>
>> On Jun 16, 2004, at 9:10 AM, Michael Latta wrote:
>>
>>> While latency may be an issue I do not think that would be the issue
>>> here:
>>>
>>> 1) The CPU usage is only about 2-5%.
>>> 2) My G4 does not have DDR memory.
>>> 3) My G5 has 4x the RAM of the G4 (4GB rather than 1GB).  So far more
>>> can be cached in ram.
>>>
>>> Does this mean that Self has trouble making good use of L2 cache like
>>> most heap based memory models?
>>>
>>> I got part way through instrumenting the time spent per frame and 1/3
>>> was in drawing and 2/3 in the activity related methods before that 
>>> (in
>>> doStep).  I hope to get more detailed measurements later today when I
>>> try again.  I crashed the UI last time and need to start from 
>>> scratch.
>>>
>>> Do you know the latency issue is resolved in the Rev B units, or you
>>> are just saying that it is a known issue in Rev A?
>>>
>>> Michael
>>>
>>>
>>>
>>> On Jun 15, 2004, at 10:57 PM, Steve Dekorte wrote:
>>>
>>>>
>>>> On Jun 15, 2004, at 3:03 PM, Michael Latta wrote:
>>>>> If I use 120 it starts to slow down, but anything less gets the 
>>>>> same
>>>>> 8-10 fps.  The times are very consistent between two very different
>>>>> machines: 1Ghz G4 and 2.0 Ghz G5.
>>>>
>>>> Apple's first generation G5 has much higher latency memory than the
>>>> G4,
>>>> so I wouldn't expect Self to be much faster. When the G5 came out, I
>>>> ran compile tests on a dual 2 Ghz G5 and they were no faster than on
>>>> my
>>>> dual 1.25 Ghz G4.
>>>>
>>>> -- Steve
>>>>
>>>>
>>>>
>>>>
>>>> Yahoo! Groups Links
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>
>>>
>>> Yahoo! Groups Links
>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>>
>>
>> Yahoo! Groups Links
>>
>>
>>
>>
>>
>
>
>
>
> Yahoo! Groups Links
>
>
>
>
>




More information about the Self-interest mailing list