[volunteergrid2-l] Revisiting space contribution caps

[Shawn Willden]

On Thu, Feb 3, 2011 at 9:43 AM, Jody Harris <jharris at harrisdev.com> wrote:

> It's important to keep in mind that the amount of usable data per user is
> limited by the minimum node size and the number of nodes, not the amount of
> space a user contributes.
>
> S = Min Node Size
> N = Number Of Nodes
> U = Number Of Users
> E = Expansion Factor (7:15 - 2.2)
> A = space available per user
>
> Something in the range of:
>
> A = (S * N)/(U * E)
>
> .... I think... and that gives us some buffer space.
>

Hmm.

I prefer not to fix E to any value, or even any range of values.  Encoding
parameter choice is an individual decision -- and eventually I'd like to
vary it based on the kind of file I'm storing, though Tahoe can't do that
very well yet.  Still, you can just omit that and consider "A" to the be
space after expansion, which simplifies your expression to:

A = (S * N) / U

So, what does that do if we allow arbitrarily-large nodes?  Suppose we have
a grid of 19 500 GB nodes and one 10.5 TB node.  So, 20 TB total, split 20
ways gives a per-user size of 1 TB.  If everyone tried to use their 1 TB, it
wouldn't work, because after 500 GB * 21 = 10.5 TB has been stored, all but
the big node is full.  It still has 10 TB free, but it's not usable.

Even if you limit each user to the smaller of A and the size of their node
(min(A,node_size)), the grid could fill up before everyone used their fair
share -- though I think this is pretty unlikely.

-- 
Shawn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tahoe-lafs.org/cgi-bin/mailman/private/volunteergrid2-l/attachments/20110203/d7083478/attachment.html>