On Wed, 22 Apr 2009 10:34:09 -0600
Shawn Willden <[email protected]> wrote:

> What about using the permuted list to find a set of servers that know
> the set of servers that hold the shares?  The share location list
> deployed to the first n servers in the permuted server list would be
> small, so it wouldn't impose a large burden either on servers storing
> the list or on clients uploading the list, and it would allow
> complete freedom to place shares wherever the client wants them.

Interesting. I had been thinking about putting a "other-share server hint
list" on each share-holding server, as a bit of metadata next to the share.
Repairers/rebalancers would be obligated to update it, and the server itself
could verify the contents every once in a while (and also use the other
shares to verify its own).

But putting merely the share-server list on the permuted-list servers (wow,
try saying that 10 times fast) might have some better properties. Or combine
the two approaches. You'd only need to get one of the first-order servers to
find the probable location of all of the second-order servers, and finding
any of the second-order servers would give you a strong possibility of
finding the rest of the second-order servers.

If the server-selection algorithm is trying to keep all your download traffic
local (whether at your mom's house or in a single colo), it could pick the
most convenient/closest first-order server to get the list from, then choose
whichever of the second-order servers met the requirements.

interesting...
 -Brian
_______________________________________________
tahoe-dev mailing list
[email protected]
http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev

Reply via email to