March 2017 Summaries

The blog post by Tomer Sandler discusses how to significantly speed up full table scans on ScyllaDB, a database known for handling large-scale data. By shifting from a traditional serial full table scan, which processes 42,110 rows per second, to an efficient parallel full table scan, ScyllaDB achieves a remarkable rate of 510,752 rows per second, a 12-fold increase. The post explains how ScyllaDB's architecture can leverage server and core parallelism to optimize performance by utilizing the token function for partitioning data, allowing for multiple concurrent queries across token ranges. The setup described includes a ScyllaDB cluster deployed on Google Compute Engine with specific hardware configurations, and the use of a Golang client with the gocql driver to execute the parallel scans. Key strategies for maximizing parallelism involve calculating the number of parallel queries based on nodes and cores, dividing the token range into many smaller ranges, and randomizing query execution to ensure that all shards are actively engaged. The detailed guide includes a demonstration and suggestions for configuring the scanning process, emphasizing the importance of optimizing queries to improve efficiency and throughput.

ScyllaDB Enterprise v2017-01 marks the first production-ready release of ScyllaDB's enterprise edition, offering a range of enhanced features and services built on the open-source ScyllaDB project, specifically version 1.6.1. This release includes a commercial license and provides users with tested and certified binaries, software updates, hot fixes, and 24/7 access to technical support and ScyllaDB engineers. ScyllaDB Enterprise also introduces new migration tools, enabling zero downtime migration from any Cassandra version, and prioritizes bug escalation for enterprise customers. While the open-source version of ScyllaDB offers limited support, the enterprise edition ensures long-term support for each release, maintaining its commitment to ongoing development and feature additions such as counters and materialized views.

ScyllaDB has announced a new round of funding aimed at expanding its database offerings, growing its open-source community, and enhancing features for its upcoming ScyllaDB Enterprise release. The company emphasizes its database's scalability and performance, which combines the power of Cassandra with the speed of Redis, making it attractive for users migrating from various platforms like Redis, HBase, and Aerospike. The funding will support further development of ScyllaDB's features, including workload conditioning, hot-cache load balancing, and improvements in counter support, materialized views, and secondary indexes. Future plans include focusing on core database enhancements, vertical integrations with technologies like Spark, and horizontal growth using the Seastar engine. ScyllaDB aims to lead in NoSQL database performance and availability by continuing to innovate and attract top talent while engaging with its expanding user community.

ScyllaDB 1.6.2 is a bugfix release for the ScyllaDB 1.6 stable branch, offering backward compatibility and support for rolling upgrades. Users upgrading from version 1.5.x are advised to review the 1.6 release notes due to changes in some metrics. The release addresses several issues, including errors during sstable flushes that could cause ScyllaDB to exit, problems with authentication settings affecting the system_auth keyspace, and potential memory allocation failures during scanning and flush operations, particularly when running with cache disabled. Additionally, it resolves streaming failures with compressed files and unexpected exits during the repair process. Users are encouraged to report any problems encountered.

ScyllaDB 1.5.2, a bugfix release for the ScyllaDB 1.5 stable branch, has been announced, with the recommendation to upgrade to the more recent ScyllaDB 1.6.2 for enhanced stability. This release is backward compatible and supports rolling upgrades, though users upgrading from version 1.4.x are advised to consult the ScyllaDB 1.5 release notes due to changes in metrics. The update addresses several critical bugs, including issues where ScyllaDB might abort on disk space depletion, unexpected exits during sstable flush errors, and memory allocation problems despite sufficient memory availability. Additional fixes include resolving incorrect disk size values reported by nodetool and addressing streaming failures with compressed files, as well as preventing unexpected exits during the repair process.

The performance report compares ScyllaDB and Apache Cassandra on low-end hardware, revealing that ScyllaDB consistently outperforms Cassandra in various scenarios, despite the hardware constraints. ScyllaDB, a reimplementation of Apache Cassandra, is designed to be hardware-friendly and highly scalable, achieving up to 10 times better performance due to its efficient architecture. In tests involving data loading, read throughput, and latency under different conditions and workloads, ScyllaDB demonstrated superior performance, particularly in handling data ingestion and maintaining lower latency without creating compaction backlogs. Even when both databases were tested under disk-bound conditions, ScyllaDB showed a 68% advantage during concurrent read and write operations. These results highlight ScyllaDB's ability to maintain consistent performance and low latencies, outperforming Apache Cassandra even on constrained, low-end hardware configurations, and emphasize ScyllaDB's self-tuning capabilities that reduce the need for manual configuration adjustments.