Association for Computing Machinery
Cloud datacenter applications such as web search, retail, advertising, and recommendation systems, etc., generate a diverse mix of short and long flows that carry widely varying deadlines due to their soft-real time nature. In this paper, the authors present MCP, a novel distributed and reactive transport protocol for Data Center Networks (DCNs) to achieve minimal per-packet delay while providing guaranteed transmission rates to meet flow deadlines. To design MCP, they first formulate a stochastic packet delay minimization problem with constraints on deadline completion and network stability. By solving this problem, they derive an optimal congestion window update function which establishes the theoretical foundation for MCP.