The order-preserving partitioner code (not hash-based anymore) is up now at https://issues.apache.org/jira/browse/CASSANDRA-3.
-Jonathan On Wed, Mar 11, 2009 at 6:48 PM, Jonathan Ellis <[email protected]> wrote: > Use Random for now. The OPHF is the same as the old one, i.e., not > actually OP. :) > > I'm pretty convinced at this point that it's impossible to have an > order-preserving hash that doesn't either (a) impose a relatively > short key length past which no partitioning is done (i.e., all keys w/ > the same prefix go to the same node) or is (b) very sensitive to key > length such that the keys with a given length N will not be evenly > distributed across all nodes. Or both. > > So I am working on migrating from pluggable hash functions key -> > BigInteger, to pluggable partitioning algorithms key -> EndPoint. > Without the requirement to transform to a numeric value first I think > I can create an order-preserving distribution that performs well. (I > need this for range queries.) > > So far I have just laid the foundation, here: > https://issues.apache.org/jira/browse/CASSANDRA-3 > > I hope to finish the rest tomorrow. > > -Jonathan > > On Wed, Mar 11, 2009 at 5:28 PM, Jiansheng Huang <[email protected]> > wrote: >> >> Which one is better to use? The default is Random. >> >> In Avinash's annoucement mail, we have >> (1) Ability to switch between a random hash and a OPHF. We still have the >> old (wrong) OPHF in there. I will update it to the corrected one tomorrow. >> >> Is correct OPHF in? Thanks. >> >
