[tahoe-dev] Analysis of file reliability, error reduction hack:
Josh Wilcox
wilcoxjg at gmail.com
Mon Sep 17 02:37:48 UTC 2007
For a binomial r.v. "X" where:
p = The probability of success
k = The number of successes under consideration
n = The total number of trials
P{X = k + 1} = [p/(1-p)]*[(n-k)/(k+1)]*P{X = k}
Using this relation one can calculate the probability of e.g. an N-K
erasure coded file on a network with servers whose individual reliabilities
(i.e. probability of
availability) are independently "p".
Interestingly it requires no use of choose functions,
and a single use of floating points that are raised to large
powers, so the error term should be quite small, relative to the naive
calculation. I wrote an ugly
function that calculates the relevant Cumulative Distribution Function.
Perhaps I should cut-n-
paste the monster here?
Would it be pedantic to go through calculating the
prob. and erasure coded file is available?
Tersely: Start with P{X = 0} and work from there.
Then use 1 - P{file unavailable}.
--Cheers
arc
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/pipermail/tahoe-dev/attachments/20070916/5d9f945c/attachment.html>
More information about the tahoe-dev
mailing list