Sei sulla pagina 1di 2

A Scalable Routing and Admission Control Model in

SDN-based Networks
M. Rasih Celenlioglu

H. Ali Mantar

Department of Computer Engineering


Gebze Institude of Technology
Gebze, Kocaeli, Turkey

Department of Computer Engineering


Gebze Institude of Technology
Gebze, Kocaeli, Turkey

mcelenlioglu@gyte.edu.tr

hamantar@gyte.edu.tr
a. Physical Network

ABSTRACT

ISS1

In this paper, we propose a scalable routing and admission


control model for Software Defined Networks (SDN). We use
pre-established multi-path (PMP) model to increase routing
scalability and reduce admission control time.

ISS2

Categories and Subject Descriptors

ISS1

C.2.1 [Computer Communication Networks]: Network


Architecture and Design

ISS2

Keywords
Software defined networking; controller design; QoS routing

1.

INTRODUCTION

Recently, the demand for access to the Internet has changed.


The Internet is switched to Internet of Services (IoS). The
amount of traffic in the Internet has been grown dramatically. Thus, a better network management is required. Software Defined Networking (SDN) [3] has been proposed for
efficient and flexible resource management. In SDN, control
plane and data plane are separated. Control plane is unified
in a logically central entity called controller. It has global
network state information and responsible for routing and
management of the underlying network. The data plane resides within switches. These switches forward packets based
on instructions in their flow tables. Open Networking Foundation (ONF) [3] develops OpenFlow (OF) protocol, which
provides signaling between controller and switches.
SDN has several issues to be solved. First of all, controller
becomes bottleneck as the network size increases because
it is the only entity that makes decisions on behalf of all
switches. Secondly, routing itself is a heavy process. Performing routing on behalf of all the switches will be much
heavier. Thirdly, since controller is the only decision maker,
processing delay must be very low for scalability. Finally,
controller is the only entity that is able configure flow tables
of switches. When a decision is made, controller updates

Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the Owner/Author. Copyright is held by the owner/author(s).
ANCS14, Oct 2021, 2014, Los Angeles, CA, USA.
ACM 978-1-4503-2839-5/14/10.
http://dx.doi.org/10.1145/2658260.2661770.

ESS1

b. Abstracted Network
ESS1

Figure 1: Illustration of physical network (a) and


abstracted network with multi-path (b).

flow tables. This produces high volume of signaling traffic


between controller and switches. Signaling traffic increases
as the number of switches in the network increases.
Most of the SDN related studies in the literature has been
made for data center networks. In these studies, incoming
requests to data centers are assigned to servers by controller
[4] to improve load balancing. For the scalability problems,
multiple controllers are used [1]. In [2], we proposed Bandwidth Broker based model for intra and inter-domain networks. However, signaling protocol in that model differs
from SDN.
In this paper, we propose a scalable routing and admission
control model for SDN-based networks. In our model, complex network is abstracted into a simple network that consists of ingress-egress pairs. Several paths are pre-established
(PMP) between each pair. Controller performs routing, admission control, signaling and resource management based
on these paths. Network abstraction is made for the ease of
network management and resource utilization.

2.

SDN-BASED ROUTING AND ADMISSION


CONTROL MODEL

We propose a scalable routing and admission control model


based on PMPs between each ingress switches (ISS) and
egress switches (ESS) as in Figure 1. Core switches between ISS and ESS only perform forwarding. Throughout
this work, we assume that paths are pre-defined. Controller
performs routing, admission control, signaling and resource
management based on these paths.
Figure 2 shows the functional design of controller. Domain
topology module handles events related to network topology.
Link state database stores up-to-date link state information

Admission Control

50

Domain
Topology

Pre-established
Path Database

Pre-established
Path Computation

Path
Resizing

Link State
Database

cost equalization

45

Path Load
Balancing

40
35
30
25
Path1
Path2
Path3
Path4

20

Figure 2: Functional decomposition of controller.

15
0

similar to OSPF table. PMP computation module establishes several maximally disjoint paths between each ISSESS pair (IEP) [5]. The capacity of a path is randomly
chosen at the first stage. Path Load Balancing module
provides load balancing by shifting flows from one path to
another. Controller resizes capacity of paths using PathResizing Module based on PMP loads. Controller does not
run routing algorithm online. It performs admission control
by looking in PMP database. Thus, it makes decisions fast.
Scalable routing and admission control are performed based
on PMP between each IEP. Controller does not compute
path finding algorithm for each flow request but uses PMPs.
These paths are pre-established and can be resized based
on aggregated traffic load. Whenever a new flow request
is received by an ISS, it informs the controller. Controller
finds the corresponding IEP by looking at the destination IP
address. Admission control is performed by querying PMP
database. If there is enough amount of bandwidth in corresponding IEP, the request is accepted. After acceptance,
the flow is assigned to an available path by its IP prefix.
Controller configures core switches at path establishment
step or whenever PMPs are changed. These switches take
marked packets and perform forwarding based on instructions matching with path-id in their flow table. Controller
does not modify core switches for each request. Since signaling occurs only between controller and associated ISS per request, signaling scalability is achieved. Controller performs
load balancing between paths of an IEP and path resizing
between different IEPs so as to increase resource utilization.
For load balancing, since a path consists of links, controller
finds link costs by dividing its traffic load by its capacity at
first. Then, link with maximum cost becomes path cost. After that, average path cost is computed for a pair. Finally,
controller computes the amount of flows to be shifted from
congested paths to less congested ones. Controller performs
these actions iteratively until the path costs are equalized.
In our model, controller reserves certain amount of capacity
for each path at the beginning. Only the controller is in
charge of updating path capacities. When the traffic load of
a PMPs between an IEP exceeds a certain threshold (e.g.,
95% of its maximum capacity), controller run path resizing
algorithm. The crucial point here is that path resizing process is not triggered based on individual flow requests. It is
triggered based on aggregated traffic in PMPs.
We built a simple virtual network as illustrated in Figure 1 using Mininet, Floodlight and and OF v1.0. Realtime network traffic is generated between end points. Static
PMPs are created using VLAN. Each VLAN id indicates a
single path. Figure 3 depicts that path costs are equalized
over time when controller performs load balancing periodically. We compared our model with non-SDN model that

1000

2000

3000
iteration

4000

5000

Figure 3: The cost equalization of paths.


does not perform load balancing and resizing. If only one
path is highly loaded (over 90 % of its capacity), our model
and non-SDN model accepts 57% and 11% of incoming requests respectively. When overall network is highly loaded,
our model accepts 40% and non-SDN model accepts 32 %
of incoming requests. Finally, we observed that admission
control time takes 1 ms in average when non-SDN model
varies between 13-26 ms.

3.

CONCLUSION

In this work, we propose an SDN-based routing and admission control model. Controller performs admission control based on PMPs. Network abstraction, load balancing
and path resizing methods are developed to increase network resource utilization. Signaling scalability is achieved
by resizing PMPs based on aggregated traffic in offline.

Acknowledgements
This research was supported by TUBITAK under Grant No.
113E253.

4.

REFERENCES

[1] Y. Hu, W. Wang, X. Gong, X. Que, and S. Cheng.


Balanceflow: Controller load balancing for openflow
networks. In Cloud Computing and Intelligent Systems
(CCIS), 2012 IEEE 2nd International Conference on,
volume 02, pages 780785, Oct 2012.
[2] H. Mantar, J. Hwang, I. Okumus, and S. Chapin. A
scalable model for interbandwidth broker resource
reservation and provisioning. Selected Areas in
Communications, IEEE Journal on, 22(10):20192034,
Dec 2004.
[3] Open Networking Foundation. [Online]. Available:
http://opennetworking.org/.
[4] R. Wang, D. Butnariu, and J. Rexford. Openflow-based
server load balancing gone wild. In Proceedings of the
11th USENIX Conference on Hot Topics in
Management of Internet, Cloud, and Enterprise
Networks and Services, Hot-ICE11, pages 1212,
Berkeley, CA, USA, 2011.
[5] J. Whalen and J. Kenney. Finding maximal link disjoint
paths in a multigraph. In Global Telecommunications
Conference, 1990, and Exhibition. Communications:
Connecting the Future, GLOBECOM 90., IEEE,
pages 470474 vol.1, Dec 1990.

Potrebbero piacerti anche