Das ist der Versuch einer Übersetzung des Freenet Statusberichts von Ian Clarke vom 18.2.2004:
Hi ihr, ich will nur jeden über das Geschehene auf das Laufende bringen und um Unterstützung bitten.
Die letzten paar Wochen waren sehr interessant. Seit ca. Juli 2003 hat sich die Performance von Freenet unerklärlicherweise verschlechtert; und seitdem versuchen wir herauszufinden warum dem so ist und wir versuchen eine Lösungen zu finden um seine vorherige gesunde Performance wieder herzustellen.
Letzten Endes entschieden wir, dass die wahrscheinlichste Erklärung einfach die Tatsache war, dass die Benutzer von Freenet mehr Anfragen in das System schickten, als es behandeln konnte. Das jeder Knoten der durch die Überladung arbeitsunfähig wurde nur noch die Last auf die anderen Knoten erhöhte resultierte in einer Endlosschleife; bis letzten Endes der Hauptteil der Nachrichten (oft über 95%), die über das Frrenet-Netzwerk verschickt wurden Knoten waren, die anderen Knoten von ihrer Überlastung erzählten.
Nachdem ein paar Versuche das Problem durch kleine Änderungen am existierenden Lastverteilung-Algorithmus zu lösen fehlschlugen, wurde uns schließlich klar, dass wir zu unseren ersten Prinzipien zurück mussten und eine Problemlösung ganz neu zu entwickeln.
Die Idee hinter dieser Lösung ist, dass, damit der Lastausgleichs-Algorithmus erfolgreich sein kann, er die Fähigkeit besitzen muss die Anzahl der Anfragen die das Netzwerk an einer Stelle betreten an der Quelle limitieren; und zwar muss die Anzahl der Anfragen auf ein Level reduziert werden, die das Netzwerk effizient behandeln kann.
The eventual solution involved nodes adaptively estimating how many
requests they can comfortably handle, their “request quota”, and then
intelligently allocating this quota to the nodes that want to send it
requests for data.
We devised an algorithm that could allocate this quota in such a way
that if a node wasn't using all of its quota, that this spare quota
could then be reallocated to other nodes that are more likely to use it.
This allows the total number of requests to remain under-quota, but
without being so-far under-quota that the nodes would be under-utilized.
A simple version of this algorithm was tested two weeks ago, and managed
to reduce the number of rejected requests from over 95% to less than 5%.
We have now implemented a more sophisticated version of the algorithm
and are in the process of testing and tweaking it, but early results are
very positive.
We believe that this “rate limiting” scheme might be so effective as to
have applications well beyond Freenet, and my intention is to write a
paper on the subject just as soon as we have gained sufficient
experience of how it operates in the real world. I think this
demonstrates that while the research nature of this project can
sometimes cause frustration when things don't quite go to plan, it also
allows us to develop and explore cutting-edge ideas that have
applications well beyond the specific goals of the Freenet project.
At the very core of the project's ability to maintain this momentum,
even when things aren't looking very rosy from the users' perspective,
is Matthew Toseland – the project's grossly underpaid and extremely
hardworking and talented full-time developer. Matthew asks for only
Ł850 per month, which is way below what he could earn as a commercial
software developer, and his full-time development efforts, in addition
to their direct effect, also serve as a catalyst for the contributions
of other volunteers.
Unfortunately, due to the recent poor performance, which we are now
hopefully climbing out of, we have been unable to do a formal release of
the Freenet software. We are now optimistic that, if rate limiting
continues to be as promising as it appears to be, we will be in a
position to release Freenet 0.6, the first major release in well over a
year.
The problem is that this is still 6 weeks away at an absolute minimum,
and as things stand right now, and despite the generous donations of
many Freenet users, the project is almost out of funds with which to
employ Matthew.
For this reason I am appealing to supporters of the project to, once
again, dig deep and if you can, make a contribution to the project in
the form of a Paypal donation, a Paypal subscription, or through E-Gold
if Paypal is not your cup of tea.
You can make a donation through the donations page on our website at:
http://freenetproject.org/index.php?page=donations
If, for whatever reason, you would like to make a donation through other
means, please contact me directly at ian-RdDMkVZAZeuJnvDnx1genLhSVCA9We0C@public.gmane.org
Many thanks,
Ian Clarke,
Coordinator, The Freenet Project.
