[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: problems with the aci range too small?



> Date: Tue, 17 Dec 2002 10:48:28 -0500
> From: Roger Dingledine <arma@mit.edu>
> 
> On Tue, Dec 17, 2002 at 10:27:14AM -0500, Paul Syverson wrote:
> > Mat is correct about what the OR papers say. And, I think that is
> > essentially the same as what Roger is suggesting.
> 
> Er. Essentially the same qualitatively; but we can handle more circuits
> my way. :)
>
OK I don't get that. The first bit determines which half of the
name space you are in and the rest is the same as before. What am
I missing?
 
> > The only issue I can think of as a potential problem is router twins.
> > If the aci space is allocated for each virtual router, then we would
> > need another bit or two to indicate which actual router is assigning
> > the name. But, I don't think that is (or should be for this and other
> > reasons) how router twins are handled.
> 
> Hm. Currently there would be no problem -- when the circuit is created
> it chooses which of the twins to use, and after choosing it chooses an
> appropriate aci for that connection (if you're talking to two twins then
> you have two connections open).
> 
> Even if later we want to do spread-spectrum things (eg spitting packets
> for a single conversation over several twins) or duplicate circuits for
> extra reliability, I think the right answer is to build each circuit
> separately with its own aci, and the reconstruction at the exit node
> would be at a higher level.
> 
> So I don't think twins will be a problem here.
> 
Right, pretty much what I was thinking. I just wanted to raise it in
case I was wrong about how things are now implemented
or if I had overlooked some issue.

-Paul