September 2015 Summaries

ScyllaDB, a high-performance database, was tested for its capacity to handle operations at a rate exceeding one million per second, utilizing metrics collection and monitoring through the collectd protocol and Riemann. The database is structured into shards, corresponding to CPU cores, each managing its resources independently to maximize parallel processing. During tests, ScyllaDB's unique memory management system, which bypasses the system's page cache, was highlighted, using two allocators: the Seastar allocator and the log-structured allocator (LSA). The database demonstrated efficient cache interactions, with performance metrics illustrating high cache hit rates during a read workload and significant data merging during a write workload, resulting in over 1.3 million writes per second. The testing setup employed cassandra-stress tools to simulate workloads, revealing ScyllaDB's capability to handle intensive read and write operations while maintaining memory efficiency and high throughput, as visualized through various system graphs and flame graphs for CPU analysis.