| |
|
Research Overview
Over the years, my research has evolved along two
main
research thrusts: Scalable Internet Protocols and Systems
and Real-Time Embedded Systems and Networks. In the past, I
have also dabbled in other areas, ranging from parallel computing to
robotics!
-
Scalable Internet Protocols and Systems:
My interest in this subject stems from the eminent need for scalable services
to match the explosive growth in availability and use of the Internet. My
work in this area is conducted within
the WING research group,
which has as its goal the
improvement of Web and Internet protocols and systems based on measurement, analysis,
and careful redesign. In that regard, with collaborators, I undertook
one of the
first studies on the effectiveness of caching in the Web -- a study that
led to a
number of projects under the
Oceans research initiative, including
eager content dissemination (my
1995 papers on this topic advanced the
CDN "push" content distribution model adopted years later by the likes of Akamai,
SightPath, Sandpiper, Digital Island, and Speedera),
speculative
prefetching, and
the
characterization of Web traffic self similarity (my
1996 paper on
this with Mark Crovella is
among the most cited on that subject) and
reference
locality properties in the Web. Another one of WING's early projects was
Commonwealth, a full-scale
implementation of a number of our techniques in the context of an experimental
high-performance, low-cost distributed scalable Web server. Ideas from that
project catalyzed in a
patent on Distributed Routing, which led to a successful startup, now part
of Network Appliance Inc. As a follow-up
to that project, and in light of the unique perspective that a large number
of servers may possess about network conditions (e.g., available paths,
traffic conditions, etc.), in the Mass
Servers Project, we investigated the use of end-to-end inference techniques to
build caricatures of the underlying network cloud--caricatures that could be
used to optimize the performance of content delivery protocols through
better utilization of network
resources. Alternatively, in the ITM Project, we
investigated the use of traffic managers--special network elements
strategically placed in the Internet--to complement end-to-end approaches
through the use of novel in-network and overlay control strategies to
reducing traffic burstiness and improving network utilization. Our
investigation of traffic control approaches led us to the identification of
a major, previously unrecognized vulnerability of complex systems such as
the Internet, namely the ability of an adversary to exploit the dynamics of
adaptation mechanisms that are invariably incorporated in systems and
networks (e.g., admission control, congestion control, adaptive routing,
load balancing, power management, etc.) We coined exploits of this nature
Reduction of Quality (RoQ) attacks,
which we investigated in a number of settings. The adversarial behaviors
exemplified in RoQ could be considered extreme, in the sense that the goal
is to maximize the damage to a system. More practically, non-altruistic
behaviors are likely to be "selfish" as opposed to "adversarial". This has
led us to investigate the implication of selfish behavior in a number of
networking and distributed systems settings, and to design
protocols that
leverage (as oppose to combat) such selfish behaviors.
-
Real-Time Embedded Systems and Device (Sensor and
Actuator) Networks: My interest in this line of research stems from
the fact that for most of the applications of computing and networking
that matter (medical devices, power grids, automotive control, financial
trading systems, air-traffic control systems, robotics, and process control,
just to name a few), worst-case performance, as opposed to average
performance, is key. For such applications, building a system that minimizes
the average response time or maximizes the average throughput is useless, if
such a system ends up missing critical deadlines, for example. As it turns
out, if one focuses on worst-case (as opposed to average) performance, most
of the practices that are common in system and network design must be
revisited. To that end, the
BU Real-Time
Computation and Communication Research Group that I founded and led since 1991
focused its early work on three main areas: Real-time communication protocols (e.g.
TCP protocols over ATM networks),
real-time databases
(e.g. real-time
concurrency control and
admission
control), and
real-time programming languages and OS support
(e.g. the TRA model,
and the
Cleopatra programming environment,
statistical
scheduling, and on
load
profiling in distributed real-time systems). Two of my currently active projects in this general
area are the iBench Initiative,
and the Sensorium Project. In the
iBench Initiative, we have as our central goal the development of a
rigorous discipline for the specification, programming, and maintenance of
distributed networked applications, with a particular focus on Internet and
video sensor network applications. In the
Sensorium project, we aim
to catalyze fundamental advances in image and video computing, network
protocols, and resource management to deal with unique spatio-temporal
constraints of sensor networks in general and of video sensor networks in
particular. snBench is an example project pursued within the Sensorium
infrastructure, in which we aim to design and implement the means via
which applications may (1) locate, monitor, and query Sensor Elements (SEs)
and Computing Elelments (CEs) for services they support, and (2) initiate,
control, or otherwise use such SE and CE services. In addition to supporting
these functionalities, snBench aims to provide basic resource
virtualization and management services for QoS support, including
resource mapping and allocation to
satisfy expressive QoS constraints in a distributed setting.
|
|