Starting with the stress test, we got a sense of how good a given Cassandra setup will do under an artificially standardized load. This may or may not reflect the particular use case that you are planning to use Cassandra for. You may tweak the stress test parameters to get closer to your test case. But if needed, you should simulate a load that represents the load condition that you are expecting on Cassandra. This will give you a basis of what to look out for when tuning. It will be helpful to keep some profiling running at OS level to gauge which resource is being depleted—things such as jConsole
, nodetool cfstats
, and tpstats
, and Linux commands such as iostats
, vmstats
, top
, df
, and free
can help to look through what's getting heated up or if everything is okay. We will see these tools in more detail in Chapter 6, Managing a Cluster - Scaling, Node Repair, and Backup.
Now that we have some idea of how our system is performing and what might need some tweaking, we can go ahead...