DIRS: Dynamic Initial Rate Setting in Congestion Control for Disaggregated Storage Systems

Xiaoqian Zhang, Allen Yang, Danlin Jia, Li Wang, Mahsa Bayati, Pradeep Subedi, Xuebin Yao, Bo Sheng, Ningfang Mi

Abstract: This paper focuses on network congestion control in a disaggregated storage system. In such a system, the supporting network requires low latency and is extremely sensitive to network congestion. The existing congestion control algorithms for data center networks do not work well in our target system because of the unique characteristics of the network topology and the storage I/O workload. Motivated by the existing issues, we develop a new solution, DIRS, which dynamically sets the initial sending rate for each flow. Our scheme helps improve the effectiveness of the congestion control protocols, especially under heavy I/O traffic. It chooses an appropriate initial rate for a flow and mitigates the congestion from the beginning while not degrading the flow’s network performance.