June 2016 Summaries

In a comparative benchmark of Aerospike 3.8.2.3 and Apache Cassandra 3.5.0, Aerospike provided 14x the throughput of Cassandra while also delivering 42x lower read latency and 24x lower update latency (95th percentile). The results show that Aerospike can accommodate future growth at a lower cost compared to Cassandra.

Aerospike outperformed Apache Cassandra in a 12-hour benchmark, delivering 14x higher throughput, 42x lower read latency, and 24x lower update latency. The performance advantage of Aerospike is attributed to its ability to handle high workloads without significant variance, making it more predictable than Cassandra. This results in reduced infrastructure costs, increased scalability, and improved application development experience. Aerospike's configuration was found to be simpler and more efficient compared to Cassandra, with fewer tuning parameters required. The benchmark highlights the potential of Aerospike as a cost-effective alternative for businesses seeking scalable and high-performance databases.

Aerospike Enterprise v3.8.3 introduces Rapid Rebalance, a feature that significantly reduces migration times by up to 40x. This enhancement enables faster upgrade cycles and static configuration changes for operations teams while minimizing latencies during migration windows for supported applications. The high-level design of the algorithm involves streaming record metadata from the immigrator to the emigrator in digest order, using sequence numbers to ensure message order, batching metadata in chunks, and implementing a heuristic to prevent slower migrations due to metadata transfer. The implementation outlines the steps taken by both the immigrator and emigrator nodes during migration processes with Rapid Rebalance enabled. Overall, this feature greatly improves cluster rebalancing efficiency and reduces transaction latency impact during migrations.

The text discusses how Aerospike 3.8 improves Cross Datacenter Replication (XDR) by addressing the challenge of high network latency between data centers located thousands of miles apart. It explains that traditional methods of handling this issue, such as using multiple TCP connections or batch-based writes, have limitations and can lead to inefficiencies. Aerospike 3.8 introduces pipelining, which allows for continuous sending of write operations without waiting for responses, thereby eliminating the impact of round-trip time on throughput. The text also highlights how TCP tuning is used to optimize XDR performance and mentions that future improvements may include support for out-of-order responses. Overall, these enhancements enable Aerospike 3.8 to effectively replicate write loads across high-latency links, improving the real-world experience of users with data centers around the globe.

Aerospike 3.8 introduces Cross Datacenter Replication (XDR) that replicates a significantly higher write load across high-latency links than its earlier versions, employing pipelining to completely take the round-trip time of the link out of the throughput equation, enabling XDR to keep up with client writes even under extremely write-heavy workloads. With pipelining, Aerospike can send and receive data concurrently without waiting for responses, allowing it to handle high-bandwidth links and reducing latency. The new version also includes TCP tuning adjustments to accommodate the increased in-flight data, ensuring that the system can process a large number of concurrent requests without running into TCP limits. This improvement has real-world implications for users with write-heavy workloads and data centers around the globe, potentially leading to an increase in XDR throughput by 10x or more after upgrading from Aerospike 3.7 to 3.8.