With only 2 replicas per DC, it means you're likely writing with a consistency level of either ONE or LOCAL_ONE. Everytime you hit the problematic node, the write performance drops. All other configurations being equal, this indicates an issue with the commitlog disk on the node.
Get your sysadmin to check for any issues with the disk, perhaps there's a problem with the hardware and failure is impending. As a side note, it's probably a good time to ensure that cluster repairs are in order in case the node fails. Cheers! GOT QUESTIONS? Apache Cassandra experts from the community and DataStax have answers! Share your expertise on https://community.datastax.com/.
