Traffic Engineering in Software Defined Networking SDN

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 3 Issue 5, August 2019 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD26618 | Volume – 3 | Issue – 5 | July - August 2019 Page 1320
Traffic Engineering in Software-Defined Networking (SDN)
Aung Htein Maw
Faculty of Computer Systems and Technologies, University of Information Technology, Yangon, Myanmar
How to cite this paper: Aung Htein Maw
"Traffic Engineering in Software-Defined
Networking (SDN)" Published in
InternationalJournal
of Trend in Scientific
Research and
Development
(ijtsrd), ISSN: 2456-
6470, Volume-3 |
Issue-5, August
2019,pp.1320-1323,
https://doi.org/10.31142/ijtsrd26618
Copyright © 2019 by author(s) and
International Journalof Trendin Scientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
Commons Attribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by
/4.0)
ABSTRACT
The traditionalsingle path routing can cause imbalancedlink utilization andis
not efficient for all traffic types such as long-lived large flows. Moreover,it can
lead to low network throughput and high network latency.Trafficengineering
(TE) is a key solution to solve these problems of singlepath.Themain purpose
of TE is to optimize the network resource utilization and improve network
performance by measuring and controlling network traffic. One of the TE
approach for large flows is multipath routing which distribute traffic load
among available multiple paths. However, most of multipath solutions do not
classify traffic flows (for example elephant or mice) and do not concern the
existing delays of routes. Therefore, to be intelligent multipath routing based
on traffic types, we proposed three main folds: (1) large flow detection
approach by using sFlow analyzer in real time, (2) measuring end-to-end
delays of available paths between source node and destination node where
large flow occurred and (3) reroute thelargeflowtominimumround-triptime
delay path in order to improve network performance. Through experimental
results, our proposed method gains over 30-77% throughput improvement
over reactive forwarding application which is implemented in ONOS
controller.
KEYWORDS: multipath; sdn; traffic engineering.
INTRODUCTION
The existing traffic rerouting models implement different
strategies in the multipath forwarding mechanism. The
authors in [3] propose the routing algorithm splits the
elephant flow into mice and distributes them acrossmultiple
paths based on source routing (labelbasedforwarding)with
round-robin manner. The limitation of theirmethodis thatit
requires overhead bytes to implement policy in packet
header increases linearly with path length. The difference is
that their approach uses round-robin to split trafficloadand
our method is based on estimated delays of each path.
Hedera [4] is a flow scheduling scheme to solve the hash
collision problem of Equal Cost Multipathing (ECMP). It
reduces large flow completion time(FCT)causedbynetwork
congestion and Utilizes the path diversity of data center
network topologies. The difference is that Hedera uses per
flow statistics for large flow detection, which has poor
scalability and our method uses packet sampling. DiffFlow
[5] differentiates short flow and long flow by using a packet
sampling method. It applies ECMP to short flows and
Random Packet Spraying (RPS) method to long flows. Their
method causes packet reordering problem while
transferring each packet to random egress ports because of
different packet delivery time of available paths between
source and destination. Our proposed method can avoid
reordering problem since it is flow-basedrerouting. Another
work of traffic rerouting in [6] monitors congested path by
collecting port statistics of each switch by using Open Flow
protocol. When congestion occurs, it computes the least
loaded path and reroutes some traffic flows from the
congested path. Tiny Flow [6] presents large flow detection
and random rerouting method. Once an elephant is
identified, the edge switch adds a new rule to the flow table
and collects byte count statistics periodically.Whenthebyte
count exceeds a limit, the switch picks an alternate egress
port out of the equivalent cost paths randomly for elephant,
reinstalls the new flow entry, and resets the byte count. The
drawback of Tiny Flow is the elephantflowcollision problem
at the random egress ports of aggregate switches, resulting
in poor bandwidth utilization. In this proposal, theproposed
rerouting method is mainly based on large flow
identification and end-to-end delay estimation. As soon as
large flow is detected, the controller computes delays of
parallel multiple paths between source and destination and
reroutes the large flow to the path with the least delay path
in order to improve throughput and minimize latency.
A. Proposed Method
As the elephant flows causes congestion and makes latency
to other mice flows, differentiating elephant flow from mice
flows is important process for improving network
performance. Our proposed method uses sFlow-RT analyzer
[8] which is explained in Section 1.1 to monitor large flow in
data plane and extracts large flow information fromsFlow in
every one second.
The elephant flow information consists of
source/destination IP addresses, source/destination MAC
addresses, source/destination ports, links where large flow
occurred.
As soon as large flow is detected, available shortest paths in
terms of hop counts can be calculated between source and
destination hosts. Then, end-to-end delay (d) can be
measured for each path from set of available path lists and
IJTSRD26618

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
stored in path-delay table. Path-delay table consists of path
and set of delay values. End-to-end delay measurement
method is described in Section 1.2. The average delay (Davg
) of each path can be calculated from total delays from table
and number of probe packets (Nprobe). According to these
average delays, Minimum delay path and second minimum
delay path is chosen for TCP traffic. Here, we use different
paths (minimum and second minimum) for data forwarding
and acknowledgement to improve QoS requirements.
B. Experiment
In our test bed, we use ONOS controller (version 1.8)
because of its performance, high level abstractions and API.
ONOS is distributed system which is designed for scalability
and high availability. It serves as network operating system
with separation of control and data plane for service
provider network. Our topology is created by using Mininet
emulator (version 2.2.1) which can create virtual network
and provide hundreds and even thousands of virtual hosts,
and Open Flow (version 1.0). Two virtual machines is used
for our test bed, sFlow analyzer and Mininet topology is
running on one VM and ONOS controller is running on
another. ONOS controller ran on a laptop powered by Core
i5-5200U CPU @ 2.20GHZ with RAM 4GB.Mininet andsFlow
ran on Laptop PC powered by Core i5-5200U CPU @
2.20GHZ with RAM 4GB.
For sFlow analyzer, we configure sFlow agents on all open
vswitches in topology according to Figure
1. In this figure, interface name (if name) of switch is used
as sflow agent address and target address is sflow collector
address. As we set link bandwidth 10Mbps, sampling rate is
1 in 10 packets and polling interval is 20 seconds. Then, we
define flows in our large flow detection script as presented
in Figure 2 which are used to match packets that share
common attributes. In flow definition, a flow called pair that
captures MAC addresses, IP addresses, TCP ports, interface
indexes, and calculate bytes per second for each flow. After
that, we also define threshold which is applied to metrics.
We set threshold value 1MB in this script. When the rate
value of a flow exceeds the threshold, it will notify as
‘elephant’ flow. Our ONOS application can access JSON
output from sFlow analyzer by calling REST API:
/events/json in every one second. This REST API list top
active flows, and remove all duplicates for flows reported by
sources. In our experiment, we use iperf tool as traffic
generator for TCP flow.
Figure1. sFlow agents configuration
Figure2. Defining flow and threshold
C. Experimental Testbed and Results
In our testbed, we use ONOS controller (version 1.8) among
other kinds of SDN controllers (eg.NOX, POX, Ryu,
FloodLight) because of its performance, high
levelabstractions and API. ONOS is distributed system which
is designed for scalability and high availability. It serves as
network operating system with separation of control and
data plane for service provider network. Our topology is
created by using Mininet emulator (version 2.2.1)which can
create virtual network and provide hundreds and even
thousands of virtual hosts, and OpenFlow (version 1.0).Two
virtual machines is used for our testbed, sFlow analyzer and
Mininet topology is running on one VM and ONOS controller
is running on another.
Leaf-spine topology is based on two-tier topology which is
mostly used in data center infrastructure. The leaf layer
includes switches to provide connectivityof end devices.The
spine layer provides connectivityof leaf switches.In ourleaf-
spine testbed topology in Figure 3, we use 8 switches and 8
hosts. Bandwidths of all links in this network topology are
set 10 Mbps. ONOS controller ran on a laptop powered by
Core i5-5200U CPU @ 2.20GHZ with RAM 4GB. Mininet and
sFlow ran on Laptop PC powered by Core i5-5200U CPU @
2.20GHZ with RAM 4GB.

Figure3. Leaf-spine topology
TABLE I. Parameter Setttings
Parameter Value
Leaf-Spine Topology Leaf = 4, Spine = 4
Link Speed 10 Mbps
Large Flow Detection
Threshold
>= 1Mbit/s
Sampling Rate 1-in-10 packets
Large Flow Metrics
Src Mac, Dst Mac
Src IP, Dst IP,
Src Port, Dst Port
Figure4. Average delays in round-trip time based on
number of probe packets
According to above Figure 4, we can study that ten and less
than ten probe packets result thesimilaraveragevalues with
ping results. In twentyandabove,somesignificant difference
points can be found. This is because of processing rate for
delay calculation is directly proportional to the number of
probe packets. According to Figure 5, 6, 7 and 8, our
proposed method improve throughput 30%-77% based on
delay difference values of each path.
Figure5. Throughput for different delay (40:10:20:30)
Figure6. Transfer duration for different Delay
(40:10:20:30)
Figure7. Throughput for different delay
(40:50:80:110)
Figure8. Transfer duration for different delay
(40:50:80:110)
When there is very low latency in network, our method and
reactive forwarding method can be same result after 140KB
buffer size. If large delay difference exists between available
paths, our method can optimize the network performance
(in terms of throughput and transfer duration) more than
reactive forwarding method.
D. Conclusion
The traffic engineering method is presented in software-
defined network by emulating layer 2 topology. The
proposed method leverages an SDN infrastructure to
support delay estimation and traffic rerouting. Unlike the
traditional reactive forwarding method, our proposed
method includes: differentiation elephant flows, estimation
end-to-end delay of available paths between specifiedsource
and destination and reroute the elephant flows to the least
delay path. The objective of our proposed method is to
improve network performance by measuring and managing
traffic dynamically. From the experimental results, we
investigate that the proposedmethodoptimizesthenetwork
throughput and transfer duration than the reactive
forwarding method if there exists large delay difference
among available paths.

References
[1] O. M. E. Committee, “Software-defined networking: The
new norm for networks,” ONF White Paper, vol. 2, pp.
2–6, 2012.
[2] A. Bianco, P. Giaccone, R. Mashayekhi, M. Ullio, V.
Vercellone, “Scalability of ONOS reactive forwarding
applications in ISP networks,” Computer
Communications, vol. 102, pp. 130–138, April 2017.
[3] S. Hegde, S. G. Koolagudi, S. Bhattacharya, “Scalable and
fair forwarding of elephant and mice traffic in software
defined networks,” Computer Networks, vol. 92, pp.
330–340, December 2015.
[4] M. Al-Fares, S. Radhakrishnan, B. Raghavan, N. Huang,
Vahdat, “Hedera: Dynamic Flow Scheduling for Data
Center Networks,” In NSDI, vol. 10, pp. 19–19, April
2010.
[5] F. Carpio, A. Engelmann, A. Jukan, “Diff Flow:
Differentiating Short and Long Flows for Load
Balancing in Data Center Networks,” in Proc. IEEE
GLOBECOM, vol. 10, pp. 1–6, April 2016.
[6] M. Gholami, B. Akbari, “Congestion control in software
defined data center networks through flow rerouting,”
in Proc. IEEE ICEE, pp. 654–657, May 2015.
[7] H. Xu, B. Li, “Tiny Flow: Breaking elephants down into
mice in data center networks,” in Proc. IEEE LANMAN,
pp. 1–6, 2014.
[8] http://www.onosproject.org/.
[9] Traffic monitoring using sFlow,
http://www.sflow.org/.

Traffic Engineering in Software Defined Networking SDN

More Related Content

What's hot (19)

Similar to Traffic Engineering in Software Defined Networking SDN (20)

More from ijtsrd (20)

Recently uploaded (20)

Traffic Engineering in Software Defined Networking SDN