Cassandra's performance can be affected by any of the following:
I/O contention between reads and writes (pending compactions)
Frequent GC pauses
Disk failures
Network misbehaviors
Threads and CPU contention
I would like to emphasize that a dead node in production is better than a slow node, and it is increasingly hard to identify a slow node that is to be replaced. This chapter will try to explain the tools and methods that can be implemented to identify the performance bottlenecks and take action on them or to tune your cluster for them.
In the past few releases, Cassandra has taken a lot of steps to simplify troubleshooting performance problems as well as improve the tooling around it. The Cassandra community has increasingly added features and improvements that can allow mixed workloads without any problems. Performance tuning depends on the types of operations your cluster performs most frequently and the type of queries performed on the data. I would like to...