Combined Use of Prioritized AIMD and Flow-Based Traffic Splitting for Robust TCP Load Balancing

Abstract

Abstract. In this paper, we propose an AIMD-based TCP load balancing architecture in a backbone network where TCP flows are split between two explicitly routed paths, namely the primary and the secondary paths. We propose that primary paths have strict priority over the secondary paths with respect to packet forwarding and both paths are rate-controlled using ECN marking in the core and AIMD rate adjustment at the ingress nodes. We call this technique "prioritized AIMD". The buffers maintained at the ingress nodes for the two alternative paths help us predict the delay difference between the two paths which forms the basis for deciding on which path to forward a new-coming flow. We provide a simulation study for a large mesh network to demonstrate the efficiency of the proposed approach in terms of the average blocking rate and the average per-flow goodput.

TCP flow aware adaptive path switching in diffserv enabled MPLS networks

“…The path goodput ratio becomes smaller for high goodput s-d pairs, however this is less important since the number of flows forwarded to SP is small for high goodput s-d pairs due to RER. For the rest of the paper, we use MDRR/RER in order to mitigate the flow starvation problem instead of SPQ/RER which was used in our earlier work [22].…”

“…This paper finds its roots in our preliminary studies [22,23], but differs from those in the following:…”

“…This paper finds its roots in our preliminary studies 22, 23, but differs from those in the following: We propose MDRR scheduling at the core nodes as opposed to strict priority queuing (SPQ) to mitigate the knock‐on effect, which leads to significant performance improvement. Using ns‐2, we simulate a 12‐node network at the TCP level for validation purposes and obtain entirely new results that quantify the gain in TCP‐level goodput using the proposed approach in comparison with the shortest path only routing which is most common in the current Internet. We note that TCP flow‐level simulation experiments in the context of multi‐path routing are rare due to the difficulty of maintaining states for a very large number of TCP flows, which we were able to achieve in this study. An important shortcoming of the solutions proposed in Refs.…”

“…We note that TCP flow‐level simulation experiments in the context of multi‐path routing are rare due to the difficulty of maintaining states for a very large number of TCP flows, which we were able to achieve in this study. An important shortcoming of the solutions proposed in Refs. 22, 23 is that although the average per‐flow goodput increases with respect to the shortest path routing, there are a number of flows routed over SPs whose performance degrades due to starvation of service. The proposed MDRR scheduling technique addresses this starvation problem.…”

“…An important shortcoming of the solutions proposed in Refs. [22,23] is that although the average per-flow goodput increases with respect to the shortest path routing, there are a number of flows routed over SPs whose performance degrades due to starvation of service. The proposed MDRR scheduling technique addresses this starvation problem.…”