A Study on MPTCP for Tolerating Packet Reordering and Path Heterogeneity in Wireless Networks

© Smart Internet Lab 2016
A Study on MPTCP for Tolerating Packet
Reordering and Path Heterogeneity in
Wireless Networks
Amani Alheid, Angela Doufexi, and Dritan Kaleshi
Wireless Days 2016 (WD 2016)
Toulouse, France, on March 23-25, 2016

Presentation Outline
2
• Introduction
• Motivation
• MPTCP and Congestion Control
• Packet Reorder and Solutions
• Performance Analysis
• Conclusion
2

Introduction
• Most of the devices today are supported with various interfaces allowing the
users to use different wired/wireless access technologies (xDSL-Ethernet-
WiFi-Cellular/Cellular LTE).
3
3G
WiFi
3G
WiFi
3G
WiFi
3G
WiFi
3G
WiFi

To evaluate and compare the end-to-end
performance of different multipath TCP
(MPTCP) congestion controllers when
run in conjunction with different TCP
packet reordering recovery algorithms.
Motivation
4

Multipath-TCP (MPTCP)
• The first proposal was on 2006 and it is being
standardized on the beginning of 2013.
• It is a modification of the classical TCP that
allows end-to-end data traffic to be split across
multiple paths, whilst maintaining TCP
connections at the end points (applications) .
• Design features
o Supports unmodified applications
o Works over today’s networks
o Works whenever TCP would work
5
Internet
Network
Interface
Transport
Application
TCP/IP Model
Internet
Network
Interface
Application
TCP/IP Model
MPTCP
Subflow (TCP) | .. | Subflow (TCP)

MPTCP
Connection
Connection
Management
Address
Management
Data Plan
Sequence
Space
Subflow
level
Connection
level
Data
Sequence
Mapping
(DSM)
Data
Acknowledg
ment (ACK)
Subflow
level
Connection
level
Congestion
Control
Uncoupled
Fully
Coupled
Coupled
6
MPTCP Architecture

7
MPTCP Congestion Control Objectives
• Improve throughput
A multipath flow should perform at least as well as a single path flow would
on the best of the paths available to it.
• Do no harm
A multipath flow should not take up more capacity from any of the
resources shared by its different paths, than if it was a single flow using
only one of these paths.
• Balance congestion
A multipath flow should move as much traffic as possible off from its most
congested paths.

8
MPTCP Congestion Control
•
, Wr is CWND of path r.
, W is the summation of all CWNDs

9
Packet Reorder
• Sending data through different paths increases the possibility that
out-of-order (OOO) packets received at the destination.
• The differences in paths characteristics cause OOO:
• Bandwidth or Data Rate
• End-to-end Delay or RTT
• Packet losses or drops
3G
WiFi
• When the receiver node receives OOO packets
it will store them into OOO buffer waiting for the
packets expected to precede them.
• The OOO arrival of the data packets will create
a substantial problem for multipath TCP while
reassembling them at the connection level.

10
In-Order Transmission
TCP
Socket
TCP
Socket
Application
ReceiverSender
Application
3 2 1
45

11
Out-of-Order Transmission
TCP
Socket
TCP
Socket
Application
ReceiverSender
Application
Fast Link
Slow Link
3 1
5
??4 2

12
Packet Reorder Solutions
Four Packet Reordering recovery methods that had been proposed
for single-TCP are used in this study:
• D-SACK: depends on duplicate selective acknowledgement (D-
SACK) to detect segment reordering.
• Eifel: depends on TCP timestamp option
• TCP-DOOR: responds by disabling the congestion control
• F-RTO: works with superiors retransmitting triggered by a
timeout.

13
Research Objective
• What is the impact of OOO events on the end-to-end throughput
when using MPTCP?
• How do OOO solutions that have been proposed for single-path
TCP perform with MPTCP?
• How sensitive this performance is against the data rate
difference between the wireless paths used?

14
Evaluated Scenario and System Setup
Simulation Parameters Values
Backbone - Data rate 100 Mbps
Backbone - Delay 10ms
Packet size 1400 B
Simulation time 50 sec
First subflow Second subflow
Wireless standard PHY data rate Wireless standard PHY data rate
Similar Wireless Links
Scenario 1 11g 54 Mbps 11g 54 Mbps
Scenario 2 11b 2 Mbps 11b 2 Mbps
Dissimilar Wireless Links
Scenario 3 11g 54 Mbps 11b 2 Mbps
Scenario 4 11g 6 Mbps 11b 2 Mbps
Server
Mobile
node
Backbone
Network
Second subflow
First subflow

15
• Aggregate Throughput (gThroughput)
The aggregate throughput is defined by the summation of the
throughputs of all available paths for MPTCP connections (SFs).
• Out-Of-Order Ratio (OOO-R)
OOO-R is measured to be total number of received packets being
stored in OOO-buffer over the total number of non-duplicate received
packets
• Out-Of-Order Buffer Size
The maximum number of bytes stored in the OOO-Buffer.
Performance Metrics

16
MPTCP Performance
• TCPDOOR is the best in terms of gThroughput (up to 25Mbps
regardless of the CC of the protocol).
• D-SACK with FC-CC approaches TCP-DOOR performance in
terms of gThroughput.
Scenario 1 [both (11g) 54 Mbps].
Scenario 1

17
MPTCP OOO-R and Buffer Size
• OOO-R for both TCPDOOR and D-SACK are larger than the others.
Scenario 1 [both (11g) 54 Mbps].
Reordering
Solution
Scenario 1
OOO-R (%)
Co-CC Un-CC FC-CC
NoPR 1.3 1.8 1.3
DSACK 12.2 21.9 10.0
Eifel 2.1 10.1 1.7
TCP-DOOR 42.1 42.1 27.1
F-RTO 1.9 7.9 1.3
Scenario 1
OOO Buffer Size (KB)
Co-CC Un-CC FC-CC
5.5 6.8 5.5
30.1 30.1 76.6
345.9 386.9 351.4
299.4 438.9 336.3
57.4 56.1 23.2
• D-SACK requires less memory than the other PRs particularly the TCP-DOOR
which provides a comparable gThroughput under FC-CC.

18
MPTCP Performance
• TCP-DOOR, D-SACK, and F-RTO improve the gThroughput.
• NoPR and Eifel are still unable to utilize the bandwidth of the links.
Scenario 2 [both (11b) 2 Mbps].
Scenario 2

19
• TCPDOOR, D-SACK, and F-RTO have higher path utilization.
Scenario 2 [both (11b) 2 Mbps].
Reordering
Solution
Scenario 2
OOO-R (%)
Co-CC Un-CC FC-CC
NoPR 4.5 5.7 4.3
DSACK 27.4 37.5 17.8
Eifel 5.4 11.3 5.3
TCP-DOOR 42.1 46.4 46.4
F-RTO 22.6 36.0 12.4
Scenario 2
Co-CC Un-CC FC-CC
9.6 10.9 9.6
21.9 21.9 32.8
50.6 50.6 50.6
157.2 131.3 101.2
53.3 49.2 16.4
• D-SACK and F-RTO come equal seconds in their gThroughput performance,
they require less than half the memory space of the TCP-DOOR

20
MPTCP Performance
• DSACK provides the highest throughput.
• DSACK transmits data through one subflow only.
• NoPR is still unable to utilize the bandwidth of the links.
Scenario-3 [(11g) 54 Mbps & (11b) 2 Mbps].
Scenario 3

21
• NoPR, DSACK and Eifel have low OOO-R because the transmission was
through only one SF.
Scenario 3 [(11g) 54 Mbps & (11b) 2 Mbps].
Reordering
Solution
Scenario 3
OOO-R (%)
Co-CC Un-CC FC-CC
NoPR 3.9 5.4 3.8
DSACK 3.0 9.1 2.6
Eifel 8.8 13.2 6.3
TCP-DOOR 57.9 56.6 57.4
F-RTO 27.5 45.0 22.4
Scenario 3
Co-CC Un-CC FC-CC
10.9 10.9 10.9
23.2 140.8 21.9
65.6 67.0 65.6
113.5 166.8 131.3
56.1 49.2 42.4
• TCP-DOOR requires more memory space.

22
MPTCP Performance
• DSACK provides the highest throughput and a stable performance.
• DSACK transmits data through one subflow only.
• F-RTO and TCP-DOOR improve the aggregate throughput through both
subflows
• NoPR is still unable to utilize the bandwidth of the links. Scenario 4

23
• NoPR, DSACK and Eifel have low OOO-R because the transmission was
through only one SF.
Reordering
Solution
Scenario 4
OOO-R (%)
Co-CC Un-CC FC-CC
NoPR 4.0 5.5 3.9
DSACK 8.9 22.0 3.9
Eifel 2.5 17.2 2.4
TCP-DOOR 54.3 54.1 54.5
F-RTO 17.2 31.8 30.3
Scenario 4
Co-CC Un-CC FC-CC
8.2 8.2 8.2
15.0 27.3 9.6
20.5 21.9 20.5
102.5 160.0 218.8
41.0 67.0 71.1
• TCP-DOOR requires more memory space.

24
Conclusion
• The packet reordering solutions bring a substantial performance
improvement for MPTCP by increasing the aggregate throughput
and path utilization.
• MPTCP-DSACK is recommended for asymmetrical path networks.
• MPTCP-TCPDOOR is the best for symmetrical path networks.
• DSACK requires less memory through all simulated scenarios.
• the congestion control strategies for MPTCP do not have a
significant impact in cases when the latency is dominated by the
core network rather than the access (wireless) network.

Any Question?
www.bris.ac.uk/engineering/research/csn
{amani.alheid, dritan.kaleshi, a.doufexi}@bristol.ac.uk

A Study on MPTCP for Tolerating Packet Reordering and Path Heterogeneity in Wireless Networks

Recommended

More Related Content

What's hot (20)

Viewers also liked (10)

Similar to A Study on MPTCP for Tolerating Packet Reordering and Path Heterogeneity in Wireless Networks (20)

More from Communication Systems & Networks (17)

Recently uploaded (20)

A Study on MPTCP for Tolerating Packet Reordering and Path Heterogeneity in Wireless Networks