On Friday 02 May 2008 04:13, Evan Daniel wrote:
> On Thu, May 1, 2008 at 6:32 PM, Matthew Toseland
> <toad at amphibian.dyndns.org> wrote:
> > On Thursday 01 May 2008 16:20, sich wrote:
> >  > Ermanno Baschiera a ?crit :
> >  > > Hi,
> >  > > In my opinion a good bandwidth control system should be necessary. I
> >  > > read that at the moment it's not very accurate. I think that all
> >  > > people with low bandwidth can benefit from an accurate bandwidth
> >  > > control. I mean... think about new comers who want to give a try
> >  > > running Freenet... They keep the node up for some days... their MSN
> >  > > starts to disconnect every 5 minutes, surfing becomes slow and they
> >  > > often have to reload pages... even if they set their node's output
> >  > > bandwidth to a resonable value. I'm afraid they at last could give up
> >  > > and unistall freenet.
> >  > > I had those problems, but with the last 3-4 builds, it happens much
> >  > > less often, and I can't exclude that it could be my isp's fault 
(maybe
> >  > > throttling?) or something else, not Freenet. Anyway, an accurate
> >  > > bandwidth control cannot hurt.
> >  > >
> >  > > -Ermanno Baschiera
> >  > For me the problem is that Freenet don't use all the bandwitch
> >  > avaible... I have very good bandwitch but Freenet is only using around
> >  > 40ko/s...
> >
> >  Do you have 0% pInstantReject as well? If so, your node is accepting 
every
> >  request sent to it, yet is still not using much bandwidth (compared to 
what
> >  it could do). Which is what I find on my node when I run with a high 
bwlimit:
> >  our neighbours simply don't send us enough requests to use up all the
> >  bandwidth, even taking into account that their neighbours are probably
> >  rejecting a lot of requests, so we probably get a lot of the rejected
> >  requests due to not being backed off.
> >
> >  I don't know that there's much that can be done about this. Load limiting
> >  adapts to the average network conditions, and we can't go too much beyond
> >  that without breaking routing.
> 
> You could increase the number of peers, and thus get more traffic...

True, but:
1) We'd have to scale it by bandwidth. I just turned that off due to some 
negative testing results (much fewer non-backed-off peers, also slightly 
lower payload although that may be illusory due to low uptime).
2) If we have 50 peers per opennet node, that will give darknet-only nodes a 
significant performance penalty. Therefore there needs to be a good reason to 
do it.

We might make some progress if we had a new load management system (e.g. token 
passing), under which we could explicitly tune how much we can use 
ubernodes...
> 
> Evan Daniel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: 
<https://emu.freenetproject.org/pipermail/support/attachments/20080502/cf868611/attachment.pgp>

Reply via email to