On the impact of congestion control for Concurrent Multipath Transfer on the transport layer

Due to the resilience requirements of a steadily growing number of critical Internet services (like emergency call handling or e-commerce transactions), the deployment of multi-homed network sites becomes more and more common. Having multiple Internet access paths, it seems straightforward to utilise them simultaneously in order to improve payload throughput by so-called Concurrent Multipath Transfer (CMT). Currently, CMT extensions for the two important Internet Transport Layer protocols - Multipath-TCP (MPTCP) for TCP and CMT-SCTP for SCTP - are in the focus of IETF standardisation. A challenge - which is currently very actively discussed in the IETF context - is congestion control for these CMT protocols. Based on the idea of Resource Pooling (RP), two approaches are currently discussed in the IETF: our own approach CMT/RP for CMT-SCTP as well as the MPTCP congestion control for MPTCP. Both approaches only have been roughly tested yet - mostly in similar path setups, i.e. paths having almost the same QoS characteristics, using “their” protocol. Therefore, the goal of this paper is to compare the existing approaches. Particularly, we also analyse more challenging scenarios containing dissimilar paths, i.e. paths having differing characteristics. Our goal is to provide insight into the different approaches, to support the IETF standardisation activities on CMT.¹²³