Re: [Rubydotnet-developer] Performance

[Tim S. <ti...@ih...>]

On Fri, Oct 29, 2004 at 10:36:45PM +1300, Tim Sutherland wrote:
> On Fri, Oct 29, 2004 at 09:30:59PM +1300, Tim Sutherland wrote:
> [...]
> > The first set of numbers is from my head and may be a wrong:
> > my bridge was getting ~500 calls per second, the other two were a bit
> > faster, maybe 800 or 1500.
> [...]
> > And now for the nice numbers...
> > 
> > Using Hashtable with Equals:  ~86,000 calls per second.
> > Using GCHandle: ~58,000 calls per second.
> [...]
> 
> I think Thomas' was around 5000 or so calls per second. It was
> definitely the best (because of the other bridges' Hashtable problem).
> But even that bridge can probably be made 10x faster or more pretty
> easily, as shown by my second set of results.

I checked my actual numbers...

I ran each program with whatever number of iterations made it run in
around 10 seconds. Then I repeated with twice as many iterations,
expecting around 20 seconds. (This was not always the case.)

My original bridge was getting between 445 and 888 calls per second. (It
varied a lot depending on the number of iterations, for gc reasons?)

Adding in ``lazy DotNet::Instance initialisation'' actually made it
slower, between 258 and 409 calls per second. This surprised me since
the things we're avoiding initialising are never actually used by the
benchmark. I think it's changing the gc somehow...

When I added some debugging to print out around 20,000 lines, the bridge
was about twice as fast. Again, gc reasons I think, with the linearity
of the hashtable lookups.

Changing the Hashtable to use Equals instead of == resulted in
21,000 calls per second. Adding in lazyness gave 47,000.

Using GCHandles, but not lazy: 26,000 calls per second
Using GCHandle, and also lazy: 60,000 calls per second

(So GCHandles are actually faster, contrary to my previous post.)


[Hopefully I won't have to followup quickly to this post!]