> Niklas Bergh: > > > -----Original Message----- > > > From: [EMAIL PROTECTED] > > > [mailto:[EMAIL PROTECTED] On Behalf Of Jens > > > Skripczynski > > > Sent: den 17 december 2003 13:35 > > > To: [EMAIL PROTECTED] > > > Subject: [freenet-support] network speed > > > > > > > > > Hi, > > > > > > I wanted to as a question about network speed in general. > > > > > > How effektive does freenet use the total bandwidth assigned > > > to it ? (e.g. compared to http over TCP/IP). TCP/IP has an > > > overhead of about 10% and give http another 10%, leaving a > > > total bandwidth of approx. 80% for Data connections ([1]). > > > > Freenet doesn't use HTTP between nodes. > Yeah, I know. I wanted to compare a 'normal' TCP/IP connection > with a 'normal' Protocoll ('http' - could have been FTP, edonk...) > with the freenet protocoll. > > With my assumptions I have calculated a percentage, that is lost > by using http over tcp/ip compared to the total bandwidth. > And got that I can use 80%, thus loose 20% to routing, commands > -- overhead. > > Next I tried to do the same with freenet, with the knowledge I have > about it. > > Since I (and the others) do not have a direct connection to the host,
You might have..or not. That is unknown to your node > they must use the bandwidth of the other freenet users. Thus the ratio > compared to the http over tcp/ip must be much less. Because if I use > bandwidth from others, they will use mine, too - And most of the time > at the same ratio. > > So now I tried to figure out, how much band-width is lost and got with my > assumptions something around 20% - where you say I'm wrong. > Has somebody done a calculation - experiment figuring out, how good the > actual routing is (like take a netto traffic of 5 MB data and calculate the > brutto amount of data for routing) - next calculate how much bandwidth it > consumes for each node, the calculate the percentage between the total > bandwidth and the netto Bandwidth for data. > > Have you a better assumption for freenet ? It is very hard to make an > assumtion from 'varies very much'. Varies very much with an average of 7 hops > is more helpfull... The number of hops really boils down to the popularity of data combined with the efficency of the routing data/algorithm in your node (= base effectiveness of the actual routing algorithm+your nodes experience level). Both of these facts varies very much, hence my fuzzy description.. There has been much discussion on the devl mailing list recently on how many hops that is needed to locate data. One theory is that most data should be reachable withing 10-15 hops.. If the data is very popular this value might even be decreased down to 2-3 hops or so I assume.. If the data you are requesting happens to be data that matches your nodes specialization very well then you might find it at 0 or 1 hops even. To calculate the current average effectiveness on a specific node you could probably use TrailingBytesAttempted/SentBytes from the diags page. On my high bw, stable branch, node that figure is somwhere around 0.93 when averaged over 24 hours (7GByte/7.5GByte). /N /N _______________________________________________ Support mailing list [EMAIL PROTECTED] http://news.gmane.org/gmane.network.freenet.support
