[chord] my plans to change merkle

[Benjie Chen]

so i mentioned earlier that there is a slight efficiency issue with
the Merkle implementation, namely that it tries to iterate thru the
database on every db insert, and that this could be quite expensive.

my plan to fix/change this is to keep a fixed sized array of
inserted keys in memory, and periodically inserts them into the
merkle tree, instead of insert into the merkle tree on every store.
in the worst case, i will insert keys into the merkle tree once
every time the synchronization protocol runs.

with only 200K, i can keep 10,000 keys. so this should be okay
memory-wise.

anyone has a problem with this?

benjie


-- 
benjie chen
benjie at lcs.mit.edu