[aspectj-dev] ajc scalability (status)

["Adrian Colyer" <adrian_colyer@xxxxxxxxxx>]

A quick progress report on performance
optimizations...

This table shows performance on compiles
with 2 aspects and -emacssym specified (ie "ajc +2a -emacssym"
from previous note):

Project Size     ajc 1.1.0
  ajc 1.1.1  ajc 1.1.1
                
            CVS Head   private build
    10             x(1)  
    3          3
    100            x    
     8          7
    500            x    
    23         20
    1000          45      
 145         38
    2000         112      
 286         80
    4000         327      
 779        179



(1) I didn't measure ajc 1.1.0 in this
low range.

This shows a significant improvement
over CVS head, bringing compile times down below the level of 1.1.0. The
changes are mostly small and self-contained, which is a testament to the
soundness of Mik's basic design (it just takes a lot of profiling and measuring
to find out which changes to make!). 

Here's a comparison of with and without
emacssym options on my private build:

Project Size      vanilla
  emacssym 
                
    10               3  
       3
    100              7  
       7
    500             16    
    20 
    1000            27    
    38 
    2000            50    
    80 
    4000           102    
   179 

You can see that we are scaling much
better, though still not quite linear with respect to number of files (but
that's to be expected). 

I have one more option I want to profile,
and then I need to carefully review all changes and run all the tests etc.
before checking in. There are more optimisations possible, but I don't
want to make them at this point as they bring wider change than the contained
optimisations made so far. I should be ready by end of tomorrow...

Regards, Adrian.

-- Adrian
Adrian_Colyer@xxxxxxxxxx