Just do the math? If you add one server to four, then 20% of the keys will be assigned to the new server, which will initially not be populated with anything, so you'll get at least 20% miss-rate until it is "fully" populated.
If you instead use the old server selection, then each key has about 20% chance to be located at the same server as before, which gives you an intial miss-rate of at least 80%. /Henrik Schröder On Thu, Mar 20, 2008 at 7:07 PM, Yin Chen <[EMAIL PROTECTED]> wrote: > Hi: > Did anyone ever evaluate the performance(mainly cache miss ratio) of > the ketama? Say one have 4 servers, then one added, how much cache miss > ratio can be achieved in general? > > -- > Yin Chen > School of EE & CS, Peking University
