July 2022 Summaries

The blog post "NoSQL Summer Blockbusters: From Wrangling Rust to HA Trial by Fire" explores a series of technical challenges and solutions related to database engineering, particularly focusing on ScyllaDB's innovative approaches and real-world applications. It details ScyllaDB's choice of asynchronous direct I/O for optimal performance, the misconceptions about SSD I/O, and the use of Diskplorer for disk behavior analysis. The post also shares user experiences like Rakuten's migration from Apache Cassandra due to performance issues and Kiwi.com's resilience during a datacenter fire. Additionally, it discusses modern database design trends, highlighting the need for infrastructure evolution to meet new business demands and the adoption of Rust for high-performance applications. Finally, the post touches on the PACELC theorem's relevance in understanding distributed database design tradeoffs, emphasizing ScyllaDB's capabilities in providing high-performance, low-latency solutions.

ScyllaDB University LIVE Summer Session 2022 is a free, half-day, online training event scheduled for July 28, featuring two parallel tracks: a beginner-oriented track covering NoSQL and ScyllaDB essentials, and an advanced track focusing on strategies and best practices. Led by top engineers and architects, the event includes hands-on training, introductions to new features, and a roundtable discussion with ScyllaDB experts. Participants who complete the training will receive access to further free online learning materials and have the opportunity to earn a certification and swag. To prepare, attendees are encouraged to explore related courses on ScyllaDB University, which cover topics such as Scylla architecture, data modeling, Kubernetes deployment, and advanced ScyllaDB features.

With the increasing demand for real-time applications handling petabyte-scale data, the importance of evaluating database performance at such scales has grown significantly. ScyllaDB conducted a foundational benchmark of its high-performance, low-latency database to assess its capability to manage extensive workloads effectively. The benchmark revealed that ScyllaDB could store a 1 PB dataset using only 20 large machines, achieving 7.5 million operations per second with single-digit millisecond latency, which underscores its storage density and cost efficiency. A key feature highlighted was workload prioritization, which allows users to allocate hardware resources based on task importance, enabling efficient cluster consolidation and reduced latency for smaller, critical workloads. The benchmark results serve as a valuable reference for understanding ScyllaDB's performance improvements and offer insights into conducting similar large-scale benchmarks.

ScyllaDB V represents the fifth generation of the ScyllaDB distributed database architecture, offering significant enhancements in performance, stability, and manageability, driven by innovations for extreme scale. Central to this evolution is the integration of the Raft consensus algorithm, which provides consistent operations and transactional schema changes, eliminating schema conflicts. The introduction of a new IO scheduler reduces latency impacts during heavy data streaming and repair operations, while repair-based operations make node replacements faster and restartable. ScyllaDB V also focuses on enabling massive scalability, allowing clusters to double or triple almost instantly, which is crucial for meeting the demands of data-intensive applications in today's fast-paced technological environment. This latest version addresses the increasing pressure on databases due to the exponential growth in data usage and query numbers, ensuring high performance and low latency even at extreme scales.

ScyllaDB Open Source 5.0 introduces several significant features and enhancements aimed at improving performance, scalability, and usability for data-intensive applications. Notably, it supports AWS EC2 I4i servers with Intel Xeon "Ice Lake" processors, offering double the performance compared to the I3 series. The release includes optimizations to the I/O scheduler, faster reverse queries, and new virtual tables for easier configuration and management. Additionally, experimental features such as strong, immediate schema changes using the Raft consensus protocol and improved tombstone garbage collection enhance reliability and ease of administration. ScyllaDB, designed to address limitations of legacy NoSQL databases like Apache Cassandra and Amazon DynamoDB, continues to outperform them in terms of throughput and latency, positioning itself as a robust choice for modern database solutions. The release follows a productive development period since ScyllaDB 4.0, with numerous code commits and minor releases, and plans for future improvements like flexible topology changes and strongly-consistent tables.