Comparison of Multipath TCP and CMT-SCTP
based on Intercontinental Measurements
Martin Becke, Hakim Adhari, Erwin P. Rathgeb
University of Duisburg-Essen,
Institute for Experimental Mathematics
Ellernstraße 29, 45326 Essen, Germany
{martin.becke, hakim.adhari, rathgeb}@uni-due.de
Fu Fa, Xiong Yang, Xing Zhou
Hainan University,
College of Information Science and Technology
Renmin Avenue 58, 570228 Haikou, China
{fufa, xyang, zhouxing}@hainu.edu.cn
Abstract—The market penetration of access devices with mul-
tiple network interfaces has increased dramatically over the
last few years. As a consequence, there is a strong interest to
use all of the available interfaces concurrently to improve data
throughput. Corresponding extensions of established Transport
protocols are receiving considerable attention within research and
standardization.
Currently two approaches are in the focus of the IETF: The
Multipath TCP (MPTCP) extension for TCP and the Concurrent
Multipath Transfer extension for SCTP (CMT-SCTP). This
paper evaluates and compares implementations of these two
loadsharing protocols by using both lab measurements and
intercontinental testbed realized via the Internet between Europe
and China. The experiments show that some performance critical
aspects have not been taken into account in previous studies.
Furthermore, they show that the simple scenario with two
disjoint paths, which is typically used for evaluation, does
not sufficiently cover the real Internet environment. Based on
these insights, we highlight that the different path management
strategies of the two protocols have a significant impact on their
performance in real Internet scenarios.
1
Keywords: Multipath Transfer, Loadsharing CMT-SCTP,
MPTCP, Buffer, Congestion Control, Performance Analysis
I. INTRODUCTION AND RELATED WORK
Nowadays, the Internet is the predominant global com-
munication infrastructure, increasing day by day in number
of users as well as in diversity of services used. Network
providers obviously aim at maximizing the utilization of the
available network resources while users expect, among other
QoS criteria, optimum and stable data throughput. In addition,
it is a basic Internet policy to ensure that each user gets his fair
share of network resources. As the network layer only provides
a very simple delivery service, current transport protocols
play a major role in striking this balance. This is exemplified
by TCP with its elaborate congestion control which tries to
optimize goodput, to limit network congestion and distribute
available network capacity in a fair manner among competing
connections.
This issue is sufficiently understood for the current Internet
scenario where two endpoints are interconnected via a single
network path, and the established solutions work reasonably
well. However, multi-homing and in particular loadsharing
over multiple network paths, create novel challenges which
require significant research effort to be fully understood. These
challenges have a major impact on the design of multipath
protocols and their mechanisms.
1
Partly funded by the National Natural Science Foundation of China
(funding number 61163014)
Although loadsharing can be applied on various OSI layers,
this paper - as well as current standardization efforts - focuses
on approaches on the Transport layer since only the Transport
protocol can easily provide a common service across the
borders of provider networks [1]. Over the years, multiple
approaches were proposed such as the Reliable Multiplexing
Transport Protocol [2], Parallel TCP [3] or mTCP [4], but
none of them has been deployed in the Internet. This could be
changed by the approaches currently discussed in the context
of the IETF. Multipath TCP (MPTCP [5]) and Concurrent
Multipath Transfer for SCTP (CMT-SCTP [6]) are two ap-
proaches supporting loadsharing for end-to-end transport. Both
protocols are in an advanced stage of the standardization in
the IETF.
One major issue for multipath loadsharing in Transport
protocols is the so called ”shared bottleneck” problem [7]
arising if several paths of a multipath connection share a
common bottleneck link and compete against classical TCP
traffic - which can neither be avoided nor detected reliably. In
this case, the multipath protocol would occupy more than the
fair share of bandwidth on the bottleneck link if every path
of the multipath connection would behave like an individual
(TCP-)connection. Therefore, the idea of Resource Pooling [8]
has been adopted for both MPTCP and CMT-SCTP [6] to
provide TCP-friendliness [9] under all circumstances which
is a basic prerequisite for standardization by the IETF and
ubiquitous deployment in the Internet.
[10] describes the current MPTCP concept and most of
the design decisions. The current effort to standardize CMT-
SCTP [11] is based on ideas first published in [6] and recently
improved and extended significantly [7], [9], [12]. Although
the standardization process is, especially for MPTCP, fairly
advanced, the behaviour of these protocols is not yet fully
understood. Some simulative performance evaluations have
already been published [6], first measurements have also been
reported [7], [9], [12]–[15]. All these evaluations mainly focus
on congestion control issues and use very simple network
topologies while evaluations in real world Internet scenarios
are still missing. Our contribution is to evaluate and compare
the specific behavior of MPTCP and CMT-SCTP both in a
lab environment and in a real world intercontinental Internet
scenario. We will first show that some performance relevant
aspects of real implementations have not been modeled suffi-
ciently in simulations so far. In a next step we will highlight
that the path management concept, which is different for
MPTCP and CMT-SCTP, has significant performance impli-
cations in real Internet scenarios.