Re: curiosity

[Timothy Stack <stack@cs.utah.edu>]

> On Friday 21 September 2001 16:25, you wrote:
> 
> > > The funny thing that is happening is that the size of the matrix
> > > influences the number of packets that get sent by the sender over my 10
> > > min testing period... (the larger the matrix, the less packets are sent).
> > >
> > > Why is this happening ??
> >
> > uh, i don't understand...
> >
> 
> My measurements show that the more lines of code the Capsule class has, the 
> less packets can be sent by the sender in a constant testing period.

So, the more work the class has to do in the course of processing a
packet, the slower it runs...  that sounds about right to me...

> >The difference has to do with the separation between flows thats new in 
> >ANTS2.  Theres some extra marshalling/unmarshalling/data copying when an
> >application crosses the protection boundary in order to send a
> >capsule.  Unfortunately, we haven't gotten around to optimizing this...
> 
> At the node level, how do you associate a caosule with a given flow ??
> with a MAC ?

Each capsule has an MD5 hash associated with it that specifies what flow
and class inside that flow should process it.  Theres a file called
On_IDs.txt in ants/doc that explains this.

> I would imagine that performing a computation (like a MAC) on the capsule 
> class before sending each capsule could result in the performance variation 
> that I am experiencing...

Its done once at flow startup

> cheers
> 
> Andreas

tim stack