On Fri, Feb 4, 2011 at 11:13 AM, Dan Kuebrich <dan.kuebr...@gmail.com> wrote: > Is 2 seconds the normal "I went to disk" latency for cassandra?
Cassandra exposes metrics on a per-CF basis which indicate latency. This includes both cache hits and misses, as well as requests for rows which do not exist. It does NOT include an assortment of other latency causing things, like thrift. If you see two seconds of latency from the perspective of your application, you should compare it to the latency numbers Cassandra reports. If you are getting timed-out exceptions, that does seem relatively likely to be the cold cache "I went to disk" case, and the Cassandra latency numbers should reflect that. =Rob
