Re: [freenet-dev] Re: Inserts can be as fast as downloads

On Fri, Apr 21, 2006 at 05:22:31PM +0000, Thomas Bruderer wrote:
> > > There seems to be the missleading concept that inserts have to be slower than
> > > downloads. The argument is: it needs to go over more hops therefore it is
> > > slower. I have discussed this issue, and I think its obvious that this is not
> > > true...
> > >
>
> >
> > You are only considering the local cost. Because an insert visits 20
> > nodes instead of 7, it will hit 3 times as many nodes. This does not
> > just affect latency! It affects throughput, for the simple reason that
> > an insert causes 3 times as much load on the network. Therefore we can
> > only send 1/3rd as many inserts as requests.
>
> but I am sure we agree that each block is most likely requested more than only
> three times? I am sure the global share of bandwith is dominated with downloads
> and not with inserts, So deven if our inserts use more bandwith than a download
> once. The global balance is probably not influenced by inserts.

So what?
>
> I see we cant prioritize inserts too much, because malicous flooding would be a
> problem then. But what about a prioritzing upto 1/3 of the whole bandwith?

Right now we do prioritize them: We use a single window for both
requests and inserts. This may be a bad thing. The practical result is
that we send more inserts than are strictly justified by the number of
rejections we get.
>
> Well First of all we have to see if the Bugfix is a solution. If I get 1/3 of my
> bandwith for my inserts which would be 10K then, would be a major step onward,
> and after that we could move to other things ;)

I'm not sure how we would do that, or if it would be a good idea. I
don't think it's unreasonable to expect inserts to be 3 times slower
than requests on a real network.

-- 
Matthew J Toseland - toad@???
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.