GridDB vs. Cassandra Benchmarks pitting the two found within

That benchmark stinks of in-memory vs persisted | IO starvation.

What workload in cassandra has no scaling with more nodes?

Need add scylladb to the benchmark list

(When it stops crashing)

4 (or 12) millions seems a really small amount of data isn't it ?

The benchmark used 4M or 12M records per node. For the 32-node test, that would equate to 128M or 384M records respectively. Each record consisted of 10x100 byte fields for the total data size to be 4GB or 12GB per node (or 128GB and 384GB for an entire 32-node cluster).

GridDB's write model keeps hot data in memory up to configured storage limit but will continually write it to persistent storage at the configured checkpoint interval.

... yes ...