MATRIZ DEL PERFIL COMPETITIVO

The research work presented here has been published in several leading journals and conferences. An initial version of our work on the choice of shared versus dedicated networks, titled “Shared versus Separate Networks: The Impact of Re-provisioning” (coauthored with Roch Guerin and Kartik Hosanagar) [77], was first published in the proceedings of ReArch’09, CoNEXT, in Rome, Italy, on December 1, 2009. A more matured version of the work was published as “The Impact of Re-provisioning on the Choice of Network Infrastructures” (coauthored with Kristin Yamauchi, Roch Guerin and Kartik Hosanagar) [81] in the proceedings of the Ninth Workshop on E-Business (WEB’10), held in St. Louis, Missouri, USA, on December 11, 2010. A journal version of the work is currently under submission in one of the leading information systems journal.

The work on network technology adoption was first published as “Dynamics of Competition between Incumbent and Emerging Network Technologies” (coauthored with Youngmi Jin, Roch Guerin, Kartik Hosanagar, and Zhi-Li Zhang) [41] in the pro- ceedings of NetEcon, SIGCOMM’08, held in Seattle, USA, between August 17-22, 2008. The work was further extended to consider the impact of converters, and was published as a journal version, titled “Modeling the Dynamics of Network Technology Adoption and the Role of Converters” (coauthored with Youngmi Jin, Roch Guerin, and Kartik Hosanagar) [80] in IEEE/ACM Transactions on Networking, vol. 18(6), 2010, along with a detailed technical report in 2009 [79], available at the University of

Pennsylvania repository.

Our results on the last topic is available in the paper titled, “Functionality-rich versus Minimalist Platforms: A Two-sided Market Analysis” (coauthored with Roch Guerin and Kartik Hosanagar) [78] in 2011, is under submission at a networking conference and plans for a journal version of the work is under consideration.

Chapter 2

Network Infrastructure Choice:

Shared Versus Dedicated Networks

2.1

Introduction

Advances in network technologies have resulted in the Internet evolving from a sim- ple data network to a global communication infrastructure that carries a multiplicity of services. This integration has many obvious advantages, but combining services with disparate requirements onto a shared network can also have a cost. It often calls for the entire network to be “upgraded” with features required by only a handful of services, and at a cost that is borne by all of them. Resource sharing can also introduce com- plex interactions among services and call for tracking and trouble-shooting problems of previously little consequence, e.g.,minor routing instabilities don’t affect most data

services but can severely degrade voice or video quality.

Hence, while sharing a network across many services is often advantageous, it need not always be, and it is of interest to determine when and why this is the case or not. The question has become even more relevant with the advent of new technologies such as virtualization [69, 83], which can further facilitate the deployment of new network “slices” dedicated to an individual new service. Conversely, even in the absence of new technologies recent instances of service deployments point to a complex decision process. For example, in deploying its new U-verse TV service AT&T chose to create a dedicated network. This was in part to ensure it could be managed more easily for better reliability and for delivering higher quality video [2]. In contrast, one of its competitors, Verizon, chose to share a common fiber optic network [15] for its own voice, video, and data services. As networking and communication technology continues to improve

and more services are network-enabled, e.g., health-care, infrastructure monitoring,

surveillance, etc., the question of whether to offer this access over shared or dedicated networks will become even more important.

For instance, the emergence of green buildings results in a facilities management infrastructure that relies upon networked sensors and actuators to monitor and con- trol building operation. This can be realized either by piggy-backing the existing IT infrastructure of a building [10], or by creating a dedicated facilities management net- work [51]. In particular, [10] provides an example of greener schools in New York that are using a shared IT and facilities management infrastructure to reduce their peak elec-

trical usage by enabling real-time monitoring over the web. In contrast, it also cites a second example, that of a sportswear retailer, which created a dedicated ethernet back- bone for its facilities management traffic because of concerns over costs, throughput and security. Thus, neither shared nor dedicated network choice emerge as an obvious winner.

Making these decisions calls for a framework that systematically examines the trade- off between shared and dedicated network infrastructures. Although the issue has re- ceived some attention, particularly from the business press, it has seen little formal analysis. Developing a framework to evaluate the underlying trade-off is the primary motivation for this work.

In this chapter, we propose a model for offering two network services, an existing service with a known demand and a new one with uncertain demand that can either be deployed on the same network as the existing service or on its own dedicated network. The model allows for economies or diseconomies of scope in network resources, and also accounts for the ability to adjust network resources (reprovision) in response to a higher than anticipated demand for the new service. The main contribution of this study is in offering a framework for service providers to evaluate network infrastructure options, and in particular to decide whether it is profitable to deploy a new service on an existing network infrastructure. The model also establishes that the extent to which reprovisioning is feasible can by itself affect which infrastructure, shared or dedicated, is more effective. In particular, two operational metrics, the gross profit margin and the

return on capacity play a major role in determining which infrastructure benefits more from reprovisioning.

The rest of the chapter is structured as follows. Section 2.2 reviews prior work from the manufacturing flexibility literature, and highlights its relevance to our research question. Section 2.3 introduces the model and its parameters. Section 2.4 presents the analysis. Section 2.6 summarizes the work’s findings and concludes the study.