Page 62 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 1

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1

management and con iguration of the telecommunication 3. STATE OF THE ART
infrastructure. Therefore, leveraging SDN to automate
This section presents an analysis of state–of–the–art
management and con iguration tasks is likely to improve
the return on investment (ROI). controller placement solutions.
This work presents a framework which can be used 3.1 Related work
by network operators to optimize their SDN controller
placement during the deployment phase. To date, there has been numerous research studies
directed towards addressing the controller placement
The paper is organized as follows: Section 2 presents problem in SDN. These can be broadly ied into
the problem being addressed by this work, Section two categories: (i) studies that implemented exhaustive
3 describes related work and their drawbacks, and algorithms, as exampled by [8] –[9] and (ii) studies that
highlights our contributions, Section 4 describes implemented heuristic algorithms, as exampled by [16]
the algorithms (based on mathematical modelling) –[30].
used to solve the controller placement problem,
Section 5 provides implementation details of the The controller placement problem was irst introduced
algorithms, Section 6 presents results from our by Heller et al. [8] in 2012. The authors study
mathematical modelling, Section 7 describes the the controller placement problem by investigating the
emulation experiments conducted to verify the outcome impact of uncapacitated controller location on average
of the mathematical modelling, Section 9 concludes the and worst–case latency. The algorithm used in this
paper. Lastly, Section 10 describes future work. study is k–center. To maintain realism, the authors
tested their algorithm on the Internet2 OS3E topology
2. PROBLEM ST ATEMENT [10]. Their results indicate that increasing the number
of controllers decreases the overall network latency
Although Local Area Networks (LANs) like Data Center with a icant trade‑off between worst–case and
Networks (DCNs) have already ited from SDN, average latency. The authors conclude that deploying
deploying SDN in real Wide Area Networks (WANs) still one controller often ices to meet existing latency
poses several design challenges. As the centralized requirements in campus networks. Expectantly, they also
brain of the network, an SDN controller must be able argue that one controller is not suf icient for large‑scale
to respond to control requests promptly. Moreover, deployments with fault tolerant requirements.
control tasks such as data‑plane monitoring, must
Hu et al. [11] proposes the use of multiple controllers
be performed as iciently as possible to maintain
to ensure reliability in the control‑plane. To optimize
up‑to‑date state information. This requires optimization
controller placement, the authors carry out a comparative
on the southbound interface. Due to the signi icant
evaluation of optimization algorithms, namely random
luence of propagation latency (switch‑to‑controller
placement, l–w greedy and brute force. They focus
latency) on WAN performance, controller placement has
their reliability metric on the ”expected percentage of
emerged as a crucial design problem that in luences
valid control paths”, where a control path is de ined
SDN’s southbound performance. Controller placement
as the interface between the switch and controller
ines the location of SDN controllers relative to
(southbound interface) as well as the connection between
the data‑plane elements, that yields better network
perfomance. controllers (east/westbound interface). The algorithms
were evaluated on Internet2 topology as well as various
Another aspect to controller placement has to do with ISP topologies from the Rockefuel database [12]. From
the number of controllers deployed in a given WAN. their simulations, random placement produced the least
Deploying a certain number of controllers has an impact optimal results, while brute force produced optimal
on several objectives such as propagation latency and results after a icantly long runtime. As a result,
reliability. Even though the number of controllers may the authors recommend the l–w greedy as the most
be known in advance, the location of these controllers optimal solution. This work is similar to Gong et al.
usually needs to be optimized to meet user requirements [13] in that they both aim to optimize reliability in
and contraints. the event of node or link failure. However, latency
(both switch–to–controller and inter–controller latency)
Therefore, the overall problem that must be addressed and load balancing are not considered in these research
is: given a real SDN‑enabled WAN, how many SDN works. Moreover, the number of controllers is assumed
controllers are needed and where should they go to to be known in advance.
optimize ined requirements and constraints
while maintaining an acceptable runtime and accuracy. Tanha et al. [9] study the controller placement problem
This is a multi‑objective optimization problem and to optimize network resilience in the event of controller
constitutes competing objectives. It is necessary to
address this problem during the early stages of SDN
planning.

46 © International Telecommunication Union, 2021