An architect is performing initial capacity planning for a new Instana self-hosted deployment that will monitor 1,000 hosts at peak load. Which resource type most commonly becomes the primary performance bottleneck and should be provisioned most generously in the Cassandra cluster at this scale?

At scale, Cassandra performance is primarily constrained by available RAM (used for OS page cache and Cassandra heap to minimize disk reads) and disk I/O throughput, as the database sustains continuous high-frequency metric writes, read operations, and background compactions simultaneously. Instana sizing guidelines emphasize provisioning ample RAM and fast NVMe SSD storage for Cassandra nodes at 1,000-host scale to prevent I/O saturation from becoming a bottleneck. CPU and network are important secondary considerations but rarely the first constraint encountered as host counts grow. The correct answer is 'RAM and disk I/O throughput, as Cassandra relies heavily on memory for caching and fast SSDs for sustained write performance'. The incorrect options — 'CPU cores, as Cassandra performs continuous metrics aggregation and compaction requiring heavy parallel computation', 'Network bandwidth between Cassandra nodes, as replication traffic dominates all other resource demands at 1,000-host scale', 'GPU resources for accelerating the time-series compression algorithms embedded in Cassandra' — are wrong because they do not align with IBM Instana's architecture or recommended practices for this scenario. Understanding this concept is essential for the Domain 5: Planning domain of the IBM Instana Observability certification.