RDMA Congestion Control: It Is Only for the Compliant

Remote direct memory access (RDMA) networks enable low latency and low central processing unit utilization, and their widespread adoption in datacenters enables improved application performance. However, there are performance isolation concerns for RDMA deployed in a shared cloud environment. In particular, congestion control enforcement and congestion control algorithms in RDMA make the network susceptible to performance hacking attacks, which give the attacker extra bandwidth and cause severe congestion in the network. These attacks can increase short flow completion times by several orders of magnitude. We surface a fundamental tradeoff in congestion control between short flow completion time and performance isolation. We discuss this tradeoff and how existing approaches do not provide a robust solution. We also advocate that researchers incorporate performance isolation concerns into the design and evaluation of congestion control.

Duke Scholars

Author Alvin R. Lebeck Computer Science

Author Danyang Zhuo Computer Science