Enhancing Rural Connectivity With Software Defined Networks

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Enhancing Rural Connectivity with Software Defined

Networks

Shaddi Hasan Yahel Ben-David Colin Scott


UC Berkeley UC Berkeley and AirJaldi UC Berkeley
Eric Brewer Scott Shenker
UC Berkeley and Google UC Berkeley and ICSI

ABSTRACT management and support costs. In this respect, a fundamental prob-


Software-defined networks (SDNs) have simplified management of lem remains: rural wireless networks are highly variable, heteroge-
complex data center and enterprise networks. We argue that SDNs neous environments that are difficult to manage, yet RWNOs oper-
can play a similar role in rural wireless networks, especially those ate while understaffed and under severe resource constraints.
in developing regions. Operating a rural network in the developing Software-defined networking (SDN) offers a principled ap-
world means coping with unpredictability, low profit margins, and proach to managing rural wireless networks, providing opportuni-
resource constraints; the increased flexibility and simplified man- ties for making their operation simpler and more efficient and po-
agement that software-defined networks provide are a major benefit tentially enabling new business models. SDNs decouple the control
in this context. Network virtualization, also enabled by SDN, could and data forwarding tasks of a network, placing control and man-
allow rural networks to operate as infrastructure providers to exist- agement functionality into a logically centralized controller which
ing ISPs, thus enabling cooperation rather than competition with configures and monitors the state of the network’s forwarding ele-
powerful incumbent providers. ments (e.g., switches and access points) [3]. In doing so, SDNs en-
able a RWNO to decouple construction of physical infrastructure,
which must be done locally, from configuration of their network,
1. INTRODUCTION which can be done remotely. Going further, this decoupling enables
Poor and rural areas are fundamentally difficult for Internet ser- the infrastructure deployment business and the ISP business to be
vice providers to profitably serve: sparse population densities re- operated by completely separate entities. As a result, software-
duce opportunities for oversubscription, low purchasing power of defined rural wireless networks can decrease costs and lower tech-
potential customers implies small profit margins, and resource con- nical and business barriers to entry, thereby enabling profitable op-
straints make providing acceptable service quality hard. The his- eration of RWNOs and expanding access to the Internet.
tory of rural wireless networks is rife with “pilot projects” that
never reached meaningful scale or slowly fell apart when the (of-
ten US-university-affiliated) team who installed them left the area.
2. OPPORTUNITIES FOR SDN
Realizing the benefits of Internet access requires profitable Internet Decoupling skills. Rural wireless network operators perform
service providers whose customers trust them to provide reliable two core operational tasks: construction of physical wireless in-
service over the long term. This requires innovation to drive down frastructure and configuration of that infrastructure. The skill sets
the costs of network operation and enable service providers to op- required for each have little overlap, and thus a RWNO must either
erate sustainable businesses, even in the most rural and poor ar- maintain separate staff for each or provide training in both areas to
eas. Recent technical innovations such as long-distance WiFi have all their staff, both expensive options. Yet for today’s rural wireless
reduced costs at the physical layer, as have regulatory decisions operators, adjusting configuration parameters on individual routers
to allocate microwave spectrum for unlicensed use. A new class and access points is commonplace, and troubleshooting link fail-
of Internet service providers has developed as a result, which uti- ures requires understanding the full networking stack. Learning
lize point-to-point, “fixed wireless” access technology to provide how to design and manage a robust and scalable data network often
service to remote, sparsely populated areas. We refer to such an comes after initial mistakes and their associated costs.
organization as a rural wireless network operator (RWNO), also Software-defined networking enables network virtualization [1],
known as “wireless ISPs” (WISPs) or “fixed wireless broadband which allows network operators to treat their physical network as
providers”. an abstract pool of resource, specify management policy against
Yet infrastructure cost is only one component of the cost struc- this abstraction, and let the SDN controller handle the configuration
ture for rural ISPs; profitable operation depends on controlling of individual network components. In doing so they decouple the
physical network from network policy. Decoupling these tasks also
enables a rural network operator to decouple the staff responsible
for each and increase specialization among their employees.
Permission to make digital or hard copies of all or part of this work for Decoupling these tasks also enables a novel solution to this
personal or classroom use is granted without fee provided that copies are training and staffing challenge: outsourcing network management.
not made or distributed for profit or commercial advantage and that copies Specifically, given a complete global view of network state and an
bear this notice and the full citation on the first page. To copy otherwise, to abstract, logical model for the underlying physical network, con-
republish, to post on servers or to redistribute to lists, requires prior specific trol plane management can be conducted from anywhere. Archi-
permission and/or a fee.
DEV’13, January 11–12, 2013, Bangalore, India. tecturally, rural wireless network operators would run one or more
Copyright 2013 ACM 978-1-4503-1856-3/13/01 ...$15.00. SDN controllers within their own network, but the policy descrip-
tion for those controllers would be crafted by the operator’s own than directly providing service to subscribers, the rural network op-
network administrators or by a third-party network management erator would provide infrastructure to existing telecom and Inter-
consultancy to translate business needs and service agreements into net service providers, its “client ISPs”. The RWNO would present
a logical network configuration. This decoupling also simplifies the these service providers, its clients, with a virtualized abstraction
deployment of novel network services, such as limited “free” tiers of its network as presented by the SDN controller. Crucially, these
of connectivity or support for local, non-Internet-based services, ISP clients would be able to modify their slice’s configuration with-
since only logical network elements must be configured to support out (non-programmatic) interaction with the RWNO, just as users
them. of cloud-hosted virtual machines require no interaction with their
Standardization of tools. A second opportunity for software- hosting provider to deploy new services. While the RWNO would
defined networking in RWNOs arises from the fragmented ecosys- still be responsible for building the physical infrastructure to con-
tem of currently available tools. RWNOs require a suite of tools nect their own network to potential clients, the subscriber would
for network monitoring, configuration, billing, and user authenti- interact directly with the client ISP for billing and support. This
cation. Complicating the situation, tools from different vendors do model of service provision is also beneficial to consumers as it en-
not necessarily interoperate or expose common configuration inter- ables multiple Internet service providers to utilize the same physi-
faces. A similar situation exists for debugging and troubleshooting: cal infrastructure, thus increasing competition.
with no unified or automated mechanisms for reasoning about the We acknowledge that the RWNO environment also poses new
status of the whole network, operators are forced to rely on ad-hoc challenges for software-defined networks, which have historically
techniques for identifying and fixing faults. Individual RWNOs de- been designed for datacenter and enterprise environments. In par-
velop institutional knowledge to cope with this situation over time ticular, production SDNs often use physically separate networks
through experience, often learned the hard way. for control and data plane traffic, and datacenter and enterprise en-
The status quo for RWNOs today is that best practices are en- vironments have few resource constraints. The loosely-coupled,
coded in their employees, and the implementation of a best prac- unreliable, and resource-constrained nature of an RWNO, cou-
tice in a network is specific to its particular environment, preclud- pled with the impracticality of building separate control networks,
ing sharing hard-earned wisdom between organizations. A solution complicates the SDN controller’s task of maintaining a consistent
naturally arises in an SDN: the controller presents a global view global view of network state. Addressing this issue is a substantial
of network state, a well-defined API and programming model for research challenge and is beyond the scope of this paper. However,
accessing and modifying that state, and implicitly a standard ab- this need not be a fatal flaw for SDN in a RWNO; indeed, it sug-
straction for monitoring and managing equipment from multiple gests a tradeoff space between the quality of the RWNO’s network
providers. Any applications built against such a controller could be and the richness of the global state the SDN controller has to work
easily shared among RWNOs. with. If the only changes to global state occur when a RWNO’s ten-
ant ISPs add, remove, or change a customer’s configuration, it may
3. VIRTUAL RWNOS be acceptable for such changes to require seconds or even minutes
before being reflected in both the controller and the affected net-
The opportunities for SDN described in Section 2 are each prac-
work data plane elements.
tical innovations that would directly impact rural wireless network
operators as they build and manage their networks today. Yet the
decoupling between construction and configuration that SDN pro- 4. CONCLUSION
vides also enables new business models for RWNOs. In particular, Just like operators of large data center networks in which SDN
these two tasks can be conducted by completely separate entities. has traditionally been applied, RWNOs are not well served by the
The logical extension of this idea is that RWNOs would, rather status quo for network management. Consolidating control and
than acting as an ISP themselves, “rent out” their network to an es- management of a rural wireless network will simplify their oper-
tablished ISP. The task of the RWNO, then, becomes simply one of ation, as will decoupling the tasks of infrastructure construction
building wireless infrastructure and ensuring it can be managed by and network configuration. This decoupling further enables new
an SDN controller. This model radically changes the way RWNOs cooperative business models for rural wireless networks. Taken to-
interact with existing telecoms. Rather than competing with in- gether, we believe SDN has an important role to play in spreading
cumbent telecoms, which often have monopoly status, government sustainable, reliable Internet access to people worldwide.
subsidy, or other strong competitive advantages, RWNOs are able
to cooperate with an incumbent provider. In this arrangement, the 5. REFERENCES
RWNO provides an incumbent telecom access to new customers [1] M. Casado, T. Koponen, R. Ramanathan, and S. Shenker. Virtualizing
outside their existing service range. In return, the incumbent tele- the network forwarding plane. In Proceedings of the Workshop on
com brings their business expertise and, if applicable, regulatory Programmable Routers for Extensible Services of Tomorrow, PRESTO
’10, pages 8:1–8:6, New York, NY, USA, 2010. ACM.
licenses to the rural market. For example, billing customers with
[2] N. Hughes and S. Lonie. M-PESA: mobile money for the “unbanked”
small and irregular incomes is challenging, but large cellular ser- turning cellphones into 24-hour tellers in kenya. Innovations:
vice providers that serve remote areas have already developed pay- Technology, Governance, Globalization, 2(1-2):63–81, 2007.
ment infrastructures that cope with this reality [2]. The RWNO can [3] T. Koponen, M. Casado, N. Gude, J. Stribling, L. Poutievski, M. Zhu,
allow such a cellular service provider to offer Internet service over R. Ramanathan, Y. Iwata, H. Inoue, T. Hama, et al. Onix: A
the RWNO’s infrastructure and thus take advantage of the payment distributed control platform for large-scale production networks.
infrastructure the telecom may already have in place. The RWNO OSDI, Oct, 2010.
itself only needs to bill and interact with its ISP “customer”.
In the long term, we envision virtual rural wireless network op-
erators which rent their infrastructure to multiple ISPs rather than
only a single one. A fully virtualized RWNO would be an infras-
tructure service provider, analogous to the role that cloud providers
such as Amazon’s EC2 plays in the server hosting market. Rather

You might also like