I'm load testing our cassandra instance. I have a column family with 2300 rows, each being ~ 5MB. And I run same query which asks for same rows. With caching='keys_only" response time is adequate. But with caching='ALL', we get four times larger response time and 40-50$ of requests don't get served, error is returned.
So I'm asking if 5MB rows are too big to be using row cache? Or maybe I should do some memory tweaks?
I'm using Cassand 1.1.10, JNA enabled.