> Even with 40 machines, you shouldn't run into any issues with
> scalability from a performance perspective.  The commercial allmydata
> site had far more than that.

You know, thinking about this - why should directories not be K=1, 
M=$numservers?  This would seem to be very roughly akin to a distributed MFT?  
Surely this data doesn't consume a significant amount of storage?  For larger 
clusters, this could curve downwards - for grids with more than 3 nodes, 
M=$numservers*.75, for grids with more than 10 nodes, M=$numservers*.50, for 
grids with more than 50 nodes, M=$numservers*.25, etc.  Since dirnodes are 
vital for the accessibility of all of their leaves, doesn't it makes sense to 
make them far more available than any single leaf?

Or, alternatively, designate a new node type - a node that is specifically 
dedicated to dirnodes?

Best,
Nathan Eisenberg



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

Reply via email to