Hello M O Faruque Sarker: I'm glad you are interested in the Tahoe-LAFS project for GSoC, and I'm glad you have your own specific thoughts about what to do.
However, I don't want distributed introduction to be so sophisticated a mechanism. My current desire for Distributed Introduction is a fairly simple algorithm which is described in http://tahoe-lafs.org/trac/tahoe-lafs/ticket/68#comment:11 . That algorithm seems sufficient to eliminate the availability Single Point of Failure and is otherwise nearly as dumb as it can be. I guesstimate the the algorithm (hereafter named "the Comment 11 algorithm") would scale up to a grid at least as large as the largest Tahoe-LAFS grids so far (on the order of a hundred servers and definitely less than a thousand servers, with perhaps a thousand clients). One possible GSoC project for you would be to first implement the Comment 11 algorithm, and then if you finish that (including tests, docs, backward- and forward- compatibility analyses, etc.) you could go on to do a "bonus round" of GSoC by implementing a simulation which shows how the Comment 11 algorithm is not scalable or has other undesirable properties. Alternately, you might be interested in a completely different GSoC project instead. What do you think? And by the way, I *do* want a copy of your conference paper, please. :-) Your locality-aware distributed robot protocol sounds interesting. Regards, Zooko _______________________________________________ tahoe-dev mailing list [email protected] http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev
