Evaluation Test and Improvement of Load Balancing Algorithms of Nginx

• Published in: IEEE access Vol. 10; pp. 14311 - 14324
• Main Authors: Ma, Chen, Chi, Yuhong
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Applications programs; Clustering algorithms; Clusters; Electronic devices; Failure rates; High concurrency; IP_HASH; load balancing; Load management; Nginx; Peer-to-peer computing; Performance enhancement; Performance evaluation; Protocols; Response time; Servers; Software; Software algorithms
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3146422

Today most people can't live without electronic devices. And more and more new devices are coming to the market every day. Apps running on these devices often connect to one or several web-based server side applications, which in turn put a lot of load and management pressure on the servers and clusters serving these web applications. Technologies such as Nginx and Keepalived were invented to address the load issues faced by these high concurrency applications. This paper tested a server cluster environment based on Nginx and Keepalived, evaluated the performance of Nginx based algorithms such as WRR, IP_HASH and LEAST_CONN, and designed an optimized version of IP_HASH (named as NEW_HASH). Compared to the original IP_HASH, the NEW_HASH reduces the probability of hash collisions and improves the performance of searching back-end nodes. The test showed NEW_HASH outperforms the original method of IP_HASH, with reduced response time, lower failure rates and increased throughput. Overall, the server cluster performed better under the high load pressure using NEW_HASH.