On Tue, 2011-02-01 11:20:52 +0100, Jan-Benedict Glaw <[email protected]> wrote: > On Mon, 2011-01-31 22:32:48 -0800, Brian Warner <[email protected]> wrote: > > * fixes to upload-happiness > > I once saw an email (or an ticket entry) addressing share balancing. > If I remember correctly, the author suggested to spread shares by > giving each node a number of key/value pairs and spreading the shares > so that the number of two nodes having a common value for any key is > minimal. That bad description may be supplied by an example: > > > # Node 1 > datacenter=1 > rack=17 > HE=1 > power_supply_chain=2 > > # Node 2 > datacenter=1 > rack=17 > HE=3 > power_supply_chain=2 > > # Node 3 > datacenter=1 > rack=17 > HE=2 > power_supply_chain=1 > > > In this example, the first two nodes share a common UPS, while node 3 > is on a different one. (Other than that, they're in the same rack in > the same datacenter.) In this example, if you had to distribute two > shares, these should go to (node 1 or node 2) and node 3, but not to > node 1 and node 2, because they share three common tag/values, while > the other combination only shares two with the others. > > As said, that was another person's idea, but I cannot find a > reference. Though it would be nice to implement, because users would > gain more control over share distribution. (Would even be nice for > running multiple nodes on one machine with several filesystems.)
Doesn't anybody have a clue who suggested this scheme of choosing
nodes for share distribution? Or did the subject distract people from
reading?
MfG, JBG
--
Jan-Benedict Glaw [email protected] +49-172-7608481
Signature of: Fortschritt bedeutet, einen Schritt so zu machen,
the second : daß man den nächsten auch noch machen kann.
signature.asc
Description: Digital signature
_______________________________________________ tahoe-dev mailing list [email protected] http://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
