A study of modern cluster-based high availability database solutions

Abstract

High availability in cloud computing is a top concern which refers to keeping service operational for the maximum amount of time. In any given application and service, the high availability of a database plays a crucial role in high application availability. For high database availability, master-slave replication has been used for a long as a high availability solution. In traditional master-slave clustering technology, write queries are routed to the master server, and the data change on the master database is propagated to one or multiple slaves asynchronously or semi-synchronously. One problem here is data inconsistency in case of an ongoing transaction. Data loss can occur if the master server goes down and fails to send requests to the slaves. As slaves are not in synchrony with their master, there could be some data delays resulting in slaves having old data. In recent years, modern multi-master clustering technologies have been made available that feature synchronous replication. This report investigates high availability clustering solutions available today, their features, and compares their similarities and differences with qualitative analysis. This thesis work aims to evaluate the high availability clustering technologies based on their performance and availability. Comparison has been made with practical implementation and running benchmark tests in Openstack Alto cloud. These High availability database solutions are the Percona XtraDB Cluster and MySQL NDB Cluster. Percona XtraDB Cluster is chosen for its instrumental key characteristics, vibrant and cutting edge features such as high performance and scaling, mitigates downtime and data loss, and allows writes to perform on any node in the cluster. It uses the storage engine XtraDB, an enhancement of the InnoDB storage engine for performance improvements and compatibility to MySQL. Moreover, it is a free and open-source high availability clustering solution for MySQL high availability (Percona, 2022). MySQL NDB Cluster is another high-availability synchronous multi-master technology. It is generally available open-source, free and vibrant solution that features horizontal scaling, and automatic table partitioning and uses an NDB storage engine designed to deliver high performance and minimize downtime (MySQL, 2022c). They have been studied and investigated qualitatively and quantitatively. Their performance is evaluated by implementing the latest version of MySQL server available at present. Experiments are carried out in the Openstack cloud platform. Experimental setup for both high availability solutions and their performance is analyzed and compared with metrics throughput, response time, CPU, and Memory usage. From the results obtained, in failover scenarios, both HA solutions handled node failures by keeping service operational and data synchronized in nodes after recovering from failures, thereby maintaining consistency in data. Benchmarking results obtained for both HA setups show performance for Readonly tests better in Percona XtraDB Cluster, for updates and delete tests, the performance of MySQL NDB Cluster is better. For ReadWrite test, both setups show more or less similar performance. In case of the need for low latency requirements, MySQL NDB Cluster is an effective solution for intensive write workloads. With not much low latency requirements for read operations with high throughput, Percona XtraDB Cluster is the effective solution in such cases. The framework for comparison and methods used in this report provides insights into which high availability solution is effective for a particular application and could be helpful for picking up the right solution for their use cases.