In some of our usage scenarios, strict low-latency requirements exist, with the core demand being that P99 or maximum latency is stably controlled within a fixed threshold. This talk shares the exploration and efforts our team has made in optimizing and implementing the low-latency mode of Apache RocketMQ. Starting from the core latency bottlenecks in the RocketMQ architecture, we sort out the key links affecting end-to-end latency, including network transmission, message scheduling, and resource allocation, and elaborate on the targeted optimizations, architectural adjustments, and implementation effect verification we have carried out. Additionally, this topic will share the methodologies and practical experiences of RocketMQ full-link latency tuning, hoping to provide some help to community developers and users.

Speakers:

Xikai Wei: Senior Development Engineer at Tencent Cloud

Senior Development Engineer at Tencent Cloud, with years of experience in computing, storage, and distributed system performance optimization.