[tahoe-dev] [tahoe-lafs] #778: "shares of happiness" is the wrong measure; "servers of happiness" is better
shawn at willden.org
Sun Aug 23 06:09:07 UTC 2009
On Saturday 22 August 2009 01:41:22 pm tahoe-lafs wrote:
> I imagine that Shawn's efforts will result in a
> web page on the client node which has knobs for encoding parameters and
> displays probable reliability properties as an output, to help users make
> those decisions. Or the other way around, where the user dials in the
> reliability requirements, and is informed of the encoding strategy (and
> resulting expansion factor).
The latter is my plan, except without describing the encoding strategy.
My thinking is that the k-of-N encoding strategy is not very meaningful to
people, especially when shares may not be perfectly distributed. The k-of-N
server strategy is a little better, but has the same basic problems. Of
course, small probabilities are also not very meaningful, but I think they're
better, especially if scaled up with significant numbers of files over
significant periods of time. I also think it's probably more useful to
describe expansion factors in terms of amount of data that can be backed up,
assuming fairness and their current available storage.
So, I want to provide users the ability to choose between a few different
reliabilities, expressed as probability of losing a file per year (or maybe
per decade), each coupled to a volume of data that can be backed up, and
perhaps also a backup time estimate. I think that puts the choice in terms
that most people can relate to.
More information about the tahoe-dev