Some notes from ~7 years of running in prod below - note though that none of this matters, the only thing that matters is benchmarking your load on your own hardware. Definitely run benchmarks and figure out what works for you.

166k/s is something you CAN hit with a 3-5 node cluster with the right data model, compaction strategy, and query pattern (mostly writes), or it could take you 10-15 nodes if you're doing tons of CAS or reads or using spinning disks or have tiny hardware.

How many bytes per insert? Are they all truly inserts or updates/upserts? Are you planning on using "IF NOT EXISTS"?

Historically offheap memtables have been faster than on heap for write only workloads
Historically stcs and twcs have been faster than LCS for write only workloads
SSDs typically better than spinning
CMS over G1 unless you can afford 16GB heap

Jeff Jirsa

NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB