International Journal of Future Generation Communication and Networking
Vol. 8, No. 2 (2015), pp. 105-114
http://dx.doi.org/10.14257/ijfgcn.2015.8.2.09
ISSN: 2233-7857 IJFGCN
Copyright ⓒ 2015 SERSC
A Throughput Improved Path Selection Method Based on
Throughput Prediction Model and Available Bandwidth for
MPTCP
Xiaomin Jin
1, 2
, Yuanan Liu
1, 2
, Wenhao Fan
1, 2
, Fan Wu
1, 2
and
Hongguang Zhang
1, 2
1
School of Electronic Engineering, Beijing University of Posts and
Telecommunications, Beijing 100876, China
2
Beijing Key Laboratory of Work Safety Intelligent Monitoring, Beijing University
of Posts and Telecommunications, Beijing 100876, China
jxm@bupt.edu.cn
Abstract
Nowadays, many network devices have more than one network interface. Meanwhile,
many protocols have been proposed to utilize multiple paths simultaneously. MPTCP
(Multipath TCP) is a transport layer protocol that transfers data by multiple paths
simultaneously, which is designed to improve the throughput and increase robustness of
end to end transmission. However, when we use MPTCP, we find that MPTCP can not
improve the throughput but even reduce the throughput in some cases. In this paper, we
study this issue. Firstly, we study the effects of path characteristics for throughput of
MPTCP, and find that path characteristics especially those bad paths have a great impact
on the performance of MPTCP. Secondly, we analyze the reasons why the bad path
characteristics have a serious impact on throughput of MPTCP, and propose a path
selection method based on throughput prediction model and available bandwidth to
improve the throughput. Finally, we validate our path selection method by simulations,
and the results show that our method can improve the throughput of MPTCP.
Keywords: MPTCP, path selection, throughput, path characteristics
1. Introduction
With the development of wired and wireless network technology, more and more
network devices are equipped with multiple interfaces (Ethernet, Wi-Fi, 3G/4G, GPRS,
etc.). The standard Internet transport protocols can use only one network interface at the
same time, which makes a waste of network resources. Many CMT (Concurrent
Multipath Transport) methods and protocols have been proposed on different layers to
utilize multiple paths simultaneously, such as application layer, transport layer and
network layer. Compared with the application layer and network layer, the transport layer
is more suitable for the deployment of CMT. MPTCP is a set of extensions to regular
TCP, which enables a TCP connection to transport data through multiple paths at the
same time [2]. MPTCP is designed to be backward compatible. The existing applications
can use MPTCP to transport data without any modifications, and developers can use the
original APIs to develop new applications. MPTCP is compatible with the Internet as it
exists now, which makes it easier to be deployed.
MPTCP will increase the resilience of the connectivity by using multiple paths at the
same time, and will get higher throughput by increasing the efficiency of the network
resource usage. Some CMT protocols like [1] run regular TCP congestion control on each
path, which is easy to implement relatively and can get high throughput added by each
TCP connection theoretically. But in reality, it is not fair to standard TCP connection