Summary: | Accelerated by on-demand computing, the number and diversity of the
Internet services is increasing. Such online services often have unique
requirements for the underlying wide-area network: For instance, online
gaming service might benefit from low delay and jitter paths to client,
while online data backup service might benefit from cheaper paths.
Unfortunately, today's Internet does not accommodate fine-grained,
service-specific wide-area route control. In this dissertation, I achieve
the following goals: 1) improve the access to the routes, 2) quantify
the benefits of fine-grained route control, and 3) evaluate the
efficiency of current payment schemes for the wide-area routes.
* Improving access to wide-area route control. Online services
face significant technological and procedural hurdles in
accessing the routes: Each service in need to control the Internet
routes, has to obtain own equipment, Internet numbered resources, and
establish contracts with upstream ISPs. In this dissertation, I propose
and describe implementation and deployment of a secure and scalable
system which provides on-demand access to the Internet routes. In
setting such as cloud data center, the system can support multiple
online services, providing each service with an illusion of direct
connectivity to the neighboring Internet networks, which, for all
practical purposes, allows services to participate fully in the
Internet routing.
* Quantifying the benefits of fine-grained route control. Even
if online services are presented with wide-area route choice, it is not
clear how much tangible benefit such choice provides. Most modern Online
Service Providers (OSP) rely primarily on the content routing to
improve network performance between the clients and the replicas. In
this dissertation, I quantify the potential benefit the OSPs can gain if
they perform a joint network and content routing. Among other findings,
I find that by performing joint content and network routing, OSPs can
achieve 22% larger latency reduction than can be obtained by content
routing alone.
* Modeling and evaluating the efficiency of the current payment
schemes for wide-area routes. Finally, increasing diversity and
sophistication of the online services participating in the Internet
routing poses a challenge to payment models used in today's
Internet. Service providers today charge business customers a blended
rate: a single, "average" price for unit of bandwidth, without regard
to cost or value of individual customer's flows. In my dissertation, I
set to understand how efficient this payment model is and if more
granular payment model, accounting for the cost and value of different
flows could increase the ISP profit and the consumer surplus. I develop
an econometric demand and cost model and map three real-world ISP data
sets to it. I find that ISPs can indeed improve the economic efficiency
with just a few pricing tiers.
|