[Mnet-devel] "Experiences Deploying a Large Scale Emergent Network"(fwd)

Eugene Leitl Eugene.Leitl@lrz.uni-muenchen.de
Sun, 17 Mar 2002 22:43:52 +0100 (MET)

-- Eugen* Leitl <a href="http://leitl.org">leitl</a>
ICBMTO: N48 04'14.8'' E11 36'41.2'' http://www.leitl.org
57F9CFD3: ED90 0433 EB74 E4A9 537F CFF5 86E7 629B 57F9 CFD3

---------- Forwarded message ----------
Date: Sun, 17 Mar 2002 05:46:20 -0800
From: Zooko <zooko@zooko.com>
To: mnet-devel@lists.sourceforge.net
Subject: [Mnet-devel] "Experiences Deploying a Large Scale Emergent Network"

I gave a talk about my experiences with Mojo Nation at the First International
Workshop on Peer-to-Peer Systems [1] recently.  Here is the position paper that
accompanied the talk:

"Experiences Deploying a Large Scale Emergent Network"

While speaking in front of this small audience of eminent systems researchers,
and while answering their questions, I realized that a simple hack should make a
big difference in the overall reliability of the file space:  do not publish
data to a server unless you have seen that server before, more than one hour ago!
And more generally, use the age and average availability of a server as a
general heuristic for preferring to publish to him.  This was idea was inspired
by a DHT ("distributed hash table") design named "Kademlia" that I really like:

See my paper to understand why this heuristic would make such a big difference.

It is easy to implement something like this using a handicapper, and I intend to
do it for the next Mnet release.



[1] http://www.cs.rice.edu/Conferences/IPTPS02/
[2] http://www.cs.rice.edu/Conferences/IPTPS02/109.pdf

Security and Distributed Systems Engineering

Mnet-devel mailing list