Professional Documents
Culture Documents
net/publication/296700857
CITATIONS READS
0 1,261
4 authors:
Some of the authors of this publication are also working on these related projects:
Technology Enhanced Assessment Methods (TEAM) in Science and Health View project
All content following this page was uploaded by Diarmuid O'Briain on 04 March 2016.
Abstract 1 Introduction
Over the last ten years or so the landscape in comput-
Networking and telecommunications have been spared ing has changed dramatically with the Cloud, large-scale
the major changes that have occurred in comput- data centres and virtualisation. Over the last few years
ing over the last decade. The Information Tech- networks have increased in speed and there has been a
nology (IT) world transformed, virtualisation, then convergence on Ethernet as the standard for all links, to
the cloud, instantly adaptable and elastic comput-
the point that the difference between Local Area Network
ing. Software Defined Networking (SDN) and Net-
work Function Virtualisation (NFV) are about to
(LAN), Metropolitan Area Network (MAN) and Wide
bring about instantly adaptable and elastic network- Area Network (WAN) has diminished dramatically. What
ing. SDN is being realised in the data centre today has not changed in that time however is the core switching
and is about to take stage in the Wide Area Network and routing functions which are generally delivered on a
(WAN). hardware based stand-alone device that is self sufficient in
SDN is the extraction of the control functions from terms of the data it switches or routes and the control nec-
networking equipment hardware. This leaves the essary to make that happen. In a bid to outdo each other
hardware with only data plane functionality. The con- to maintain advantage in the market companies like Cisco,
trol plane functions are migrated as software func- Juniper and HP have loaded their devices with features
tions to be ran on standard industry hardware or that over time have resulted in network devices that rely
more often than not on server instances located on on aged protocols like Border Gateway Protocol (BGP) to
cloud platforms. communicate and networks have levels of header encapsu-
NFV is a separate but complementary technology lation that eat into the Maximum Transfer Unit (MTU)
that replaces existing functions typically found on size of the packets. This layering of abstractions on top of
specialised hardware with virtualised versions of the other abstractions is not conducive to Network Manage-
same function. These NFVs can be delivered on a ment, where traffic patterns are decided within each layer
virtual Customer Premises Equipment (vCPE) de- independently.
vices that will provide virtualisation locally for the
provision of NFVs and/or in concert with cloud based
It is not uncommon for a packet to arrive at an Inter-
functions at the data centre. net service provider (ISP) network with a Virtual LAN
(VLAN) tag, the ISP adding another VLAN tag before
This revolution will create the appearance of infi-
passing the packet to an upstream ISP who adds an Multi-
nite capacity and permit the expansion of the current
Protocol Label Switching (MPLS) header as it is switched
scientific, informatics and engineering boundaries to
create a Cloud Integrated Network (CIN). CIN, the across their IP network.
Internet of Things (IoT) as well as AuGmented In- While the underlying networks have converged towards
telligence (AuGI) will come together in the future to the all Ethernet / all Internet Protocol (IP) model, in some
create the perfect storm that will transform human form the number of services have increased rapidly. In the
existence in a third industrial revolution [1]. past ISPs provided Internet Access in the form of Broad-
Symposium on Transformative Digital Technologies - Kampala 2016
Page 2 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
tures without the limitations of the constraints of today’s • The network should be governed by policies declared
networks [6]. Funding from this initiative from 2003 until over high-level names.
2005 released researchers at a number of US Universities
• Network routing should be policy-aware.
from the boundaries created by existing Internet design
decisions while taking advantage of the benefit of hind- • The network should enforce a strong binding between
sight and the lessons already learned in a drive to develop a packet and its origin.
a 100 Mb/s to 100 million US households (100 x 100).
The outcome of the research was a system, demon-
strated in figure 3 that separated the control and data
3.1 4D Architecture planes, with a controller governed by a policy manag-
From this the idea that the Control and Data planes ing communication between end-hosts to a point where
should be separated was a theme that evolved from the there are no connections without explicit permission. It
Clean Slate 4D [7] approach which proposed an architec- also proposed specialised Ethane switches with data-paths
ture with a separate decision plane responsible for manage- managed by a flow table. Flow table entries to consist of
ment and control, a dissemination plane to control commu- a matching Header linked to a corresponding Action.
nications from the control entity to the routing devices, a
discovery plane to monitor traffic and changes within the
topology plus a data plane to handle the actual traffic.
This 4D architecture allows for the direct control of data
plane resources by an abstracted decision plane.
Page 3 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
promotion and adoption of SDN through open standards switches were interconnected with R2 [25] Channel Asso-
development. OpenFlow has evolved to version 1.5.1 [14] ciated Signalling (CAS), typically in E1 Link: G.732 [26]
(As of Apr 2015). - G.704 [27] Framing circuits. Like the Internet today the
telephone switches of that era communicated both the sig-
nalling and bearer channels over the same physical links.
6 SDN Controller development
Now that a standard SBI existed the evolution of con-
trollers as well as work on a NBI became important. Net-
work Operating System (NOX) [15] a C++ based first
generation controller [16] was developed by Nicira Net-
works and donated to the research community. A Python
version of the NOX Controller called POX [17] was de-
veloped for rapid development and prototyping. Another
Python based SDN Controller is ’RYU’ [18] (Japanese:
flow), available under the Apache 2.0 license has Open-
Stack integration and supports OpenFlow 1.0 – 1.4 plus
Nicira extensions. Ryu has a Web Server Gateway In-
terface (WSGI) and by using this function, it is possible
to create a REST API (called RESTful API) [19], which
is a useful NBI link with other systems or browsers in
an application tier. A commercial grade Java SDN Con-
troller developed by Big Switch Networks evolved from a
Java based research SDN Controller called Beacon [20] as Figure 5: Signalling System No. 7 (SS7).
Project Floodlight [21]. This project code is also Apache
2 licensed. It, like RYU, has a RESTful API. The other In the 1970s Signalling System No. 7 [28] Common
big SDN Controller is a Linux Foundation collaborative Channel Signalling (CCS) was developed to separate the
project called OpenDaylight (ODL) [22], developed in signalling from the bearer channels, this released the con-
Java. The latest version of the platform designated He- trol from the telephone switches allowing the Operators
lium is a follow on from the first release of ODL called to deliver richer centralised services known as Intelligent
Hydrogen. This project was designed to take advantage of Network (IN) [29] services. Referring to figure 5 the links
existing Linux Foundation projects, like integration with between the switches carry the bearer channels, the voice
OpenStack as well as developments with high availability, equivalent of the Data Plane and are called Inter Machine
clustering and security. ODL OpenFlow plugin supports Trunks (IMT). From a Signalling perspective the switches
OpenFlow versions 1.0 and 1.3. Like RYU and Project contain an entity called a Service Switching Point (SSP)
Floodlight an application tier is made possible through a which performs the call processing on calls by interact-
RESTful API as well as an Authentication, Authorisation ing with the connected SS7 Signal Transfer Points (STP).
and Accounting (AAA) AuthN filter. These STPs act as SS7 routing devices passing SS7 mes-
sages between SSPs, Service Control Points (SCP) and
other STPs. The SCPs offer telephony services on the In-
6.1 NBI Developments telligent Network (IN). So SS7 is in effect a Control Plane
As SDN evolves it has become apparent that new NBI while IN is a network of telephony services.
mechanisms are required to meet the diverse Applications
that will call on the SDN Controller. The Frenetic Project 8 Network Function
[23] raises the level of abstraction for programming SDNs
by the development of simple, reusable, high level abstrac- Virtualisation
tions and efficient runtime systems that automatically gen-
erate and install corresponding low-level rules on SDN At the SDN & OpenFlow World Congress in Darmstadt,
switches. Pyretic [24] is a Frenetic Project implementa- Germany in October 2012 a group of Tier 1 service
tion embedded in Python. providers launched an initiative called NFV [30]. These
operators could see that Virtualisation and Cloud comput-
ing could evolve the way services are delivered on networks
7 SS7 in the telephony industry by consolidation and virtualisation of network equipment
on industry standard high volume servers as can be seen
The changes being witnessed in the migration to SDN from in the figure 6 NFV concept. Functions could also be mi-
traditional networking is analogous to the changes in the grated to centralised virtualised infrastructure while also
telephony industry in the late 1970s and 80s. Telephony offering the facility to push virtualisation of functions right
Page 4 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
out to the end user premises. While SDN and NFV are
complimentary to each other they are not as yet inter-
dependent and can therefore be operated either together,
or independently. Obviously moving functions that were
heretofore based on specialist hardware presents a number
of challenges, such as;
• the interaction of the Management and Network Or- Figure 7 shows the overall NFV ecosystem [31]. The
chestration (MANO) of the distributed functions with underlying infrastructure collectively is called the Net-
the network. Using the benefits of automation to work Function Virtualisation Infrastructure (NFVI) and
achieve the transformational aspects of NFV. it consists of three domains, Network, Compute and Hy-
pervisor/Virtualisation. The Network Domain consists of
• the integration of functions into the overall NFV islands of switches with SDN Controllers or a traditional
ecosystem and its coexistence with legacy systems. routed and switched network. The computing hardware
and storage necessary to support the upper layers form the
• the new challenges in terms of security and stability Compute Domain consists. The final domain in the NFVI
have evolved as a result of cloud computing and vir- is the Hypervisor/Virtualisation Domain which contains
tualisation. the virtualisation hypervisors and VMs. This can be built
using existing hypervisors like KVM, Xen, VMWare or
These challenges and newer security challenges will using Container technology like Docker. These NFVI do-
evolve from this new networking system. The benefits mains are managed by a Virtual Infrastructure Manager
of NFV however make the case for migration so com- (VIM). A Virtual Network Function Manager (VNFM)
pelling that without doubt it will form the core of services controls the building of individual Virtual Network Func-
to be offered by service providers well into the future. tions (VNF) on the VMs. MANO performs the overall
Hardware-based appliances have a specific life, which is management of the VIM, VNFM and Operations Support
getting shorter and shorter with the rapid pace of devel- Systems (OSS) / Business Support System (BSS) and al-
opment, and they need regular replacement. This compli- lows the service provider to quickly deploy and scale VNF
cates maintenance procedures and customer support with services as well as provide and scale resources for VNFs.
no financial benefit to the service provider. NFV will This system reduces administrator workloads and removes
transform the design of the network to implement these the need for manual administration type tasks. It also of-
functions in software, many of these will process centrally fers APIs and other tooling extensions to integrate with
thereby allowing for their operation to be migrated and existing environments.
backed up as needed. This will reduce equipment costs
and reduce power consumption due to power management
8.1 Providing NFV to the customer
features in standard servers and storage, while eliminating
the need for specific hardware. Services can be scaled up Figure 8 demonstrates the benefits that ubiquitous high
and down in a similar fashion to that provided by cloud speed broadband gives to the service provider. It provides
services today. IT MANO mechanisms familiar today in the ability to supply a vCPE [31] to the customer upon
cloud services will facilitate the automatic installation and which VNFs can be offered. Current services that can be
scaling of capacity by building Virtual Machines (VM) or converted into NFV style services are:
Page 5 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
• Definition of interfaces.
• Router.
• Establishment of a vibrant NFV ecosystem.
• Session Border Controller (SBC).
• Performance and assurance considerations.
• Load Balancer.
• Security.
• Network Address Translation (NAT).
Page 6 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
potential of SDN. The IETF has a Software-Defined Net- increased interconnected computing world combined with
working Research Group (SDNRG) [37] with the stated AI systems exceeding human intelligence, could in fact re-
goal of identifying the approaches that can be defined, place human intelligence in a near future singularity event
deployed and used in the near term as well identifying [39].
future research challenges. The IETF have also a Net-
work Function Virtualisation Research Group (NFVRG)
(IETF, n.d.) to focus on research problems associated References
with NFV-related topics and the research community to
[1] M. K. Weldon, The Future X Network: A Bell Labs Per-
address them. The Linux Foundation believe that with the spective. CRC PressINC, 2015. 00000.
projects they have in place already, they are in a perfect
position to bring these together as a new project Open [2] N. Feamster, H. Balakrishnan, J. Rexford, A. Shaikh, and
Platform NFV (OPNFV) to accelerate NFV [35]. Dr. J. Van Der Merwe, “The case for separating routing from
James Kempf of Ericsson believes that NFV and SDN routers,” in Proceedings of the ACM SIGCOMM work-
have traversed the peak of inflated expectation and are shop on Future directions in network architecture, pp. 5–
12, ACM, 2004. 00298.
starting down the trough of despair [38]. However he has
considered the OPNFV initiative of the Linux Foundation [3] M. Caesar, D. Caldwell, N. Feamster, J. Rexford,
which he sees as a complimentary effort to their existing A. Shaikh, and J. van der Merwe, “Design and implemen-
OpenDaylight and OpenStack projects. He believes that tation of a routing control platform,” in Proceedings of
there is a lot of work yet to be achieved before reaching the the 2nd conference on Symposium on Networked Systems
slope of enlightenment and considers that SDN is confined Design & Implementation-Volume 2, pp. 15–28, USENIX
to the data centre for some time to come. Association, 2005. 00377.
Page 7 of 8
Symposium on Transformative Digital Technologies - Kampala 2016
[14] Open Networking Foundation (ONF), “OpenFlow Switch [33] I. ETSI, “Network Functions Virtualisation–Network Op-
Specification, Version 1.5.1 ( Protocol version 0x06),” erator Perspectives on Industry Progress - 2014,” Updated
Mar. 2015. 00000. White Paper, 2014. 00000.
[15] “NOX · GitHub.” 00003. [34] ETSI, “Network Functions Virtualisation.” 00008.
[16] N. Gude, T. Koponen, J. Pettit, B. Pfaff, M. Casado, [35] Linux Foundation, “OPNFV - OPNFV - An open plat-
N. McKeown, and S. Shenker, “NOX: towards an oper- form to accelerate NFV,” Oct. 2014. 00000.
ating system for networks,” ACM SIGCOMM Computer
[36] ONRC, “Open Networking Research Center (ONRC).”
Communication Review, vol. 38, no. 3, pp. 105–110, 2008.
00000.
00991.
[37] IETF, “IRTF Software-Defined Networking Research
[17] “POX · GitHub.” 00000.
Group (SDNRG).” 00000.
[18] RYU project team, “RYU SDN Framework — Ryubook
[38] J. Kempf, “NFV and SDN: Has the Hype Curve Peaked?,”
1.0 documentation.” 00000.
Jan. 2014. 00000.
[19] T. Fredrich, RESTful Service Best Practices. Pearson [39] S. Hawking, S. Russell, M. Tegmark, and F. Wilczek, “{S}
eCollege, 2012. 00006. tephen {H} awking:\’{T} ranscendence looks at the im-
plications of artificial intelligence-but are we taking {AI}
[20] D. Erickson, “The beacon openflow controller,” in Pro-
seriously enough?\’,” The Independent, vol. 2014, no. 05-
ceedings of the second ACM SIGCOMM workshop on Hot
01, p. 9313474, 2014. 00019.
topics in software defined networking, pp. 13–18, ACM,
2013. 00135.
Page 8 of 8