Network interaction has become ubiquitous with the advent of the information age and has penetrated all areas of our lives, such as Such as cloud gaming, web search and autonomous driving, to promote human progress and bring convenience to society. However, the growing number of clients also caused some problems that affected the user experience. Online services may not respond to some users within the expected time frame, known as high tail latency. In addition, the bursty traffic of the server exacerbates this problem.
To solve this problem and improve computer performance, researchers need to constantly optimize network stacks. At the same time, low entropy cloud (i.e. low interference between workloads and low system jitter) is emerging as a new trend where the Labeled Network Stack (LNS) based server is a good case to gain orders of magnitude performance improvement compared to servers achieve based on traditional network stacks. Therefore, it is imperative to perform a quantitative analysis of LNS to uncover its benefits and potential improvements.
Wenli Zhang, a researcher at the State Key Laboratory of Processors, Institute of Computing Technology and co-authors of this study, said, “Although previous experiments have shown that LNS compared to mTCP, a typical user-space network stack in academia, and Linux Network stack, the mainstream network stack in the industry, lacks a thorough quantitative study to answer the following two questions:
(i) What is the main reason for the low tail latency and low entropy of LNS compared to mTCP and Linux network stacks?
(ii) How much LNS can be further optimized?”
To answer the above questions, an analysis method based on queuing theory is proposed to facilitate the quantitative study of cloud server tail latency. In the massive client scenario, Zhang and co-authors establish models that characterize the change in processing speed in different phases for an LNS-based server, an mTCP-based server, and a Linux-based server, using burst traffic as an example. In addition, the authors derive the formulas for the tail latency of the three servers.
“Our models 1) show that two technologies in LNS, including prioritized full-datapath processing and full-path zero-copy, are primary contributors to high performance, with orders of magnitude improvement in tail latency since reducing latency entropy by a maximum of 5.5 times that of the mTCP-based server and 2) suggesting the optimal number of worker threads querying a database, reducing the concurrency of the LNS-based server by 2.1×−3 .5 × is improved. Zhang said, “The analytical method can also be applied to modeling other servers characterized as tandem-stage queuing networks.”
This work is supported in part by the National Key Research and Development Program of China (2016YFB1000200) and the National Natural Science Foundation of China Key Program (61532016).
Article Reference: Hongrui Guo, Wenli Zhang, Zishu Yu, Mingyu Chen, “Queueing-Theoretic Performance Analysis of a Low-Entropy Labeled Network Stack”, Intelligent Computing, vol. 2022, Article ID 9863054, 16 pages, 2022. https://doi.org/10.34133/2022/9863054