May 2022 Summaries

In a comprehensive benchmark test comparing ScyllaDB and Apache Cassandra, ScyllaDB demonstrated superior performance and cost-efficiency. Despite using only four high-capacity nodes compared to Cassandra's 40 smaller nodes, ScyllaDB delivered 33% higher throughput under high load conditions and was able to perform major compactions 32 times faster. The test, which involved both UPDATE and SELECT queries across a multi-terabyte dataset, showed that ScyllaDB's architecture could leverage modern hardware more effectively than Cassandra, resulting in significant cost savings and reduced administrative burden. Scaling operations with ScyllaDB were also found to be 11 times faster than with Cassandra. These results highlight ScyllaDB's ability to maintain equivalent or superior performance with fewer resources, making it a compelling option for companies seeking efficient database solutions.

Over the past decade, the development of fully distributed databases has become increasingly prominent, distinguishing themselves from traditional databases by enabling global distribution across public clouds, availability zones, and even multi-cloud and hybrid cloud environments. This shift has led to the emergence of new database systems designed specifically for distributed deployments, while some existing systems have adapted by incorporating distributed architectural components. The article explores the current landscape of distributed databases, examining both SQL and NoSQL systems, and highlights the top contenders such as PostgreSQL, CockroachDB, MongoDB, Redis, and ScyllaDB. It assesses these systems based on their clustering capabilities, replication models, and topology awareness, noting that CockroachDB and ScyllaDB provide the most comprehensive feature sets for SQL and NoSQL databases respectively. The discussion underscores the lack of a universal standard for distributed databases, emphasizing the ongoing evolution within the industry to enhance flexibility, performance, and resilience of these systems.

In the blog post, Raouf Chebri discusses five strategies to optimize CQL queries for applications using ScyllaDB, a NoSQL database known for its high availability and sub-millisecond writes. The author emphasizes the importance of using prepared statements to enhance query efficiency by caching parsed query strings, thus reducing redundant parsing. Paging queries can improve latency by limiting the amount of data scanned and retrieved, while avoiding "ALLOW FILTERING" prevents inefficient full-table scans by utilizing indexes for data lookups. Chebri also suggests using the "BYPASS CACHE" command for rare range scans to directly access data from the disk, bypassing cache lookups. Finally, selecting an appropriate Consistency Level, specifically QUORUM, balances availability and latency by requiring a majority of nodes to provide a response, ensuring data is replicated across the cluster efficiently. The ScyllaDB Monitoring Dashboard is recommended for early-stage development and testing to identify and address potential performance issues before they impact users.

AWS's new EC2 I4i instances, powered by 3rd generation Intel Xeon Scalable processors, offer significant improvements in performance for data-intensive workloads, particularly for ScyllaDB, a high-performance NoSQL database. The I4i instances feature up to 30 TB of local Nitro SSD storage, providing enhanced I/O performance and reduced latency compared to previous generations. Benchmark tests show that ScyllaDB achieved up to 2.7 times higher throughput per vCPU on I4i instances compared to I3 instances for read operations, and 2.2 times higher throughput with a 40% reduction in latency for mixed read and write operations. These instances also boast higher CPU frequencies and more memory, making them ideal for applications capable of leveraging a large number of cores, such as ScyllaDB with its shard-per-core architecture. The I4i's performance improvements are attributed to AWS's custom-designed Nitro SSDs, which have improved storage performance, resulting in impressive throughput and lower latency, even with increased workloads.

The blog post discusses the deployment of a Graph Data System using JanusGraph and ScyllaDB, as taught in a ScyllaDB University course. A graph database uses nodes and edges to represent and link data, with JanusGraph being a scalable, open-source option optimized for large volumes of vertices and edges. The lesson provides step-by-step guidance on deploying JanusGraph with ScyllaDB as the data storage layer, utilizing Docker containers and AWS infrastructure. It also highlights the performance advantages of ScyllaDB over other storage backends like Apache Cassandra and HBase, with ScyllaDB demonstrating significantly higher throughput and better query performance. The use of the Gremlin graph traversal language, part of Apache TinkerPop, allows for flexible graph traversals applicable to both transactional and analytical workloads. The post concludes with references to real-world use cases and further learning opportunities through ScyllaDB University.