Collaborative Research: RIPS Type 2: Strategic Analysis and Design of Robust and Resilient Interdependent Power and Communications Networks

Project: Research project

Description

3.3.2 PROPOSED RESEARCH Incentive Alignment Between Heterogeneous Interdependent Networks A basic idea around which the Internet architecture is designed, from physical layer schemes, routing and scheduling algorithms, to transport layer controls is that of best effort service by pushing bits from A to B as quickly and cheaply as possible. Several different competing Internet Service Providers (ISPs), both compete for revenue through carriage of these bits, as well as collaborate to ensure their end-to-end carriage across networks. As communication on the Internet becomes increasingly tied to the control of other infrastructure such as the power grid, it is becoming obvious that elements belonging to heterogeneous networks that are owned by different entities both depend on and are depended upon by other critical infrastructure elements A question faced by several ISPs today is that of the economic viability of provisioning the access network with high-speed optical links. The possibility has been considered for years \cite{wired:fiber}, but only recently has there been small-scale fiber-to-the-home in the US \cite{verizon:fiber}. The same applies to back haul provisioning of cellular base stations. This hesitation is perhaps a relic of Dot Com era fiber deployment where the value of attracting traffic was considered so high that no thought was given to whether it actually covered provisioning costs. In the same manner, utility companies, accounting for cost versus likely usage very carefully, provision transmission lines and other equipment cautiously. The question of how much to provision becomes even more important when it is clear how much value is contained in network effects, i.e., interrelationships between constituent networks. The final element in this mix of ISPs and utility companies is the 10 regulatory framework that is provided by different government agencies. Requiring certain reliability standards has been the norm in the utility industry, although the best-effort clause in the telecom business has meant that communications equipment is far less robust. This failing has been on display during events like Hurricane Sandy, when a large fraction of cellular equipment, and even wired Internet services were affected in New York City. The close linkage between different kinds of networks implies the need for a set of inter-industry standards on what reliability means, and could also necessitate government intervention to harden critical elements. With this background, our objectives are three fold, namely (i) identification of value of each infrastructure element (ii) efficient provisioning of such valuable elements, and (iii) viable revenue models for all stakeholders. We seek to answer the following fundamental questions. Provisioning Decisions: Given that each ISP or utility provider is aware of its logical dependence conditions on the other infrastructure elements, would provisioning decisions be taken that ensure reliability for the system as a whole? It seems intuitively believable that a decision framework based on knowledge about the Boolean interaction functions between networks would lead provisioning decisions under which the system as a whole is stable. However, there has been work on questions of this nature, e.g., \cite{hewal02}, in which it is shown that an ISP might actually want to reduce demand so as to attain a higher favorable return. In earlier work \cite{arun-sen}, we determined the most important network elements from the overall system perspective. However, each firm is primarily interested only in its own benefit through obtaining the services of a particular provider. If the networks were each to take provisioning decisions based on their own myopic self-interest, what would the net reliability of the system be? Would the return to each network be such that they would provision each network adequately? A hidden cost of selecting a lower provisioning value is that a network might itself be affected through the unreliability that it causes to this customer networks. Thus, value does not arise only directly through a customer-provider relationship, but also stems from the network effects of the system. But how do we find these values? This motivates the next question that we focus on. Value Determination and Incentives: What is an appropriate model for determining the levels of contribution of the different entities, and how are they to split revenues in the system? Current network architectures are assembled around the idea that communications providers carry data, while utility networks carry electricity. However, since each is integral to the availability of the other, they must be treated as partners rather than solely as consumers of each others services. This would mean incentivising them to both contribute to the value of the system through their resources, as well as to share revenue based on the perceived importance of each element in the aggregate. A game theoretical concept that defines each individual's contribution to the aggregate value is that of Shapley Value. It is calculated by calculating the difference in value of a group with and without a particular individual, and averaging over all possible combinations. The Shapley Value lends itself well to the case of Boolean relationships between network elements, since it is relatively easy to determine the value of each possible group. We propose to use this concept in order to come up with a characterization of the most important elements for the preservation of essential services. Such elements could then be hardened, either because the system as a whole recognizes their value, or because a regulatory body provides an incentive to do so. An additional aspect in the context of interacting heterogeneous networks is that of correlations across networks. Both communication networks and power networks can undergo cascading failures. Elements whose Shapley values are extremely high indicate the potential for failure cascades to propagate through them. Such elements indicate the need for topological changes to the network adding additional routes, for instance so as to educe the reliance on particular vulnerable elements. Here too, our objective will be to design efficient algorithms to compute the topological changes needed. Now, constructing good topologies requires the contracting of resources at many time scales. This motivates the following question. 11 Resource Exchange: How do we facilitate the creation of a distributed exchange that enables the trade in network resources? Since available resources at each network element are not infinite, each communications provider or utility company would have to share these resources is some manner. A basic requirement is a resource exchange at which they can trade their resources at the large and small time-scale contracting required to support their networks. While exchanges exist for utilities to buy and sell energy, there are no such equivalents where communication services can be traded. Our objective will be to design a market clearing mechanism under which each will be able to provision ``on the fly'' from available bandwidth and power resources. The scheme must be fully distributed and each agent needs to know only the price for its set of resources. One way of implementing such a scheme is through auctions of bundles of services. Each network would then bid for the services that it desires, accounting for the fact that other networks are also doing so. Under such situation, a mean filed equilibrium could arise. Here, agents assume a distribution for which their opponents actions are drawn, and construct a best response to. If the best response action is itself a sample form the assumed distribution, the system is at a mean filed equilibrium.
StatusFinished
Effective start/end date11/1/149/30/19

Funding

  • National Science Foundation (NSF): $378,000.00

Fingerprint

Internet service providers
Telecommunication networks
Heterogeneous networks
Communication
Industry
Internet
Fibers
Costs
Critical infrastructures
Optical links
Hurricanes
Routing algorithms
Scheduling algorithms
Network architecture
Base stations
Electric lines
Electricity
Display devices
Topology
Availability