[tahoe-dev] Estimating reliability

[Shawn Willden]

Has anyone done any work on code to estimate relability, given assumptions for 
k, N, peer network size and peer availability?  More to the point, has anyone 
written any code to compute recommended values for k and N, to achieve 
specified reliability for a getwork of a given size and an assumption about 
peer availability?

I'm just musing about what is needed to create a very simple configuration 
process for a backup network, and I think it would be much easier for a user 
to pick a reliability figure from a list, where each probability of loss 
value is paired with an estimate of the expansion factor.  Actually, I want 
to make it even simpler than an expansion factor.  Assuming that in a fair 
network each backup client should share as much space to others on the 
network as it consumes from them, I want to calculate how much they need to 
back up (by looking at their filesystem) and tell them "X% reliability for 
your data will require Y GB of your storage and take Z time to back up".  
With that, they can make an informed decision about X, and let the client 
then compute appropriate k and N.

If no one has done this, I'll add it to my list of things to think about.  If 
anyone has a paper/article on the calculation, that would be useful.  
Otherwise, I'll bust out my old stats texts and work it out myself.

	Shawn.