I've recently made some improvements in the SDF scheduler that result in 
much faster scheduling times for
exactly this type of model.  The scheduler previously scheduled one firing 
at a time for each actor.
Now the scheduler attempts to schedule the actor as many times as possible, 
based on the number of tokens
at its input.  This means that scheduling time is roughly O(model) instead 
of O(schedule).
Your model is scheduled essentially instantaneously with the new scheduler.

Steve

At 08:40 AM 10/31/2001 -0800, Edward A. Lee wrote:

>Steve:
>
>You did some performance tuning on the SDF scheduler in response to
>reviews, right?  Do you want to check this out?
>
>Edward
>
>>From: "Chris Cox" <[EMAIL PROTECTED]>
>>To: <[EMAIL PROTECTED]>
>>Subject: Run times using Ptolemy II
>>Date: Wed, 31 Oct 2001 10:46:30 -0000
>>X-Mailer: Microsoft Outlook Express 5.50.4133.2400
>>Sender: [EMAIL PROTECTED]
>>
>>I've been writing simulations using the SDF model that contain a mix of 
>>different data rates. If the range of data rates in the simulation is 
>>large, then the initialisation and execution times can be excessive.
>>
>>The attached file is a simple example that multiplies two data series 
>>together. One data series is up-sampled by a user-specified sample rate 
>>factor. If I run the simulation with a sample rate factor of 64 for 64 
>>iterations, then it runs very quickly. However if I run it with a sample 
>>rate factor of 4096 for 1 iteration, then it takes several seconds to 
>>initialise and execute, even though the number of multiplies should be 
>>the same.
>>
>>I've found this a big problem in more complex simulations. Can you give 
>>me any advice?
>>
>>Best regards
>>
>>Chris Cox
>
>------------
>Edward A. Lee, Professor
>518 Cory Hall, UC Berkeley, Berkeley, CA 94720
>phone: 510-642-0455, fax: 510-642-2739
>[EMAIL PROTECTED], http://ptolemy.eecs.berkeley.edu/~eal
>


----------------------------------------------------------------------------
Posted to the ptolemy-hackers mailing list.  Please send administrative
mail for this list to: [EMAIL PROTECTED]

Reply via email to