Green Computing and
Communications Laboratory
|
Current Projects
Power Efficient Scheduling for Multi-core Systems: More
computational power is offered by current embedded systems to cope with CPU
intensive applications, however, this facility comes at the price of more
energy consumption and eventually higher heat dissipation. Massively
multi-core processors are also prominently featured in the future product
roadmaps of many industry leaders.
With the birth of multi-core systems, manufacturers on one side have
introduced system with multiple cores to lead in cutting edge technology,
while on the other front, this advancement presents research community with
enormous challenges i.e., handling thermal issues and non uniform
performance of cores, and lack of mature scheduling theory, to name a few.
Maintaining performance symmetry among the cores is one of the vital issues
researchers are dealing with currently both from hardware and software
perspectives. So far, two possible solutions to non uniform frequencies of
cores are explored (i) adding dynamic voltage circuitry per core
(hardware), and (ii) schedule tasks among cores wisely to enable all cores
run at optimal uniform frequency. The first compensation strategy exhibits
leakage power at higher frequencies and undermine thermal throttling. Being
the promising alternative, the latter is relatively unexplored from
scheduling point of view as compared to uni-processor counter part. In
particular, very little is known so far for applying power-efficient
scheduling on multi-core systems. Considering this gap, we partition the
given workload among cores with the intention that all cores run at the same
uniform speed for maximum gain in energy savings.
Integration of DVS into Non-preemptive Real-time Systems:
With explosive growth in embedded industry in general and handheld devices
in particular, more and more complex applications are being built today to
attract potential customers. The most promising scheduling strategy for such
processor hungry applications is dynamic priority scheduling, which is
optimal scheduling technique from utilization view point. Similarly, these
devices are battery powered and energy is a scared resources which must be
utilized intelligently. Owing these limitations, real-time system community
is currently focusing on the dynamic voltage scaling (DVS), to extend the
operational hours of such devices. Much of the work done on integrating DVS
into real-time systems is targeting preemptive task systems mainly and very
little attention is dedicated to the non-preemptive counterpart. Hence, the
inherit drawback associated with preemptive systems is the high number of
context switching, and represents a substantial cost to the system in terms
of CPU time. In this work, we introduce a faster test to cope feasibility
analysis at run time in non-preemptive case. Once this test is validated,
the test will be extended to integrate the DVS component so that the timing
constraints of the system remain intact, while running the CPU at minimum
possible speed.
Performance-sensitive Power Aware Scheduling for Multiprocessor
Systems: Traditionally, real-time systems are focused on periodic
task models where tasks are released at regular time periods. The scheduling
algorithms that are optimal on uniprocessor are not necessarily optimal in
multiprocessor case. With maturity of multiprocessor architecture, real-time
systems today operate in dynamic environment where human activities (aperiodic
tasks) are inevitable. These tasks must be completed as soon as possible,
therefore the priority of such tasks must be higher than that of periodic
tasks. However, there is tradeoff involved between system responsiveness and
system energy consumptions. We believe, both are the main aspects of a
real-time systems, however, responsiveness becomes more important,
especially when humans interact with the system. We schedule mixed tasks
(periodic and aperiodic) on the multiprocessor platform, however, in
contrast to its counter part (uni-processor systems), the field of
scheduling mixed tasks at multiprocessor system still remains unexplored. In
this work, we are going to accommodate mixed workload with earliest-deadline
first (EDF) algorithm and proposed a restriction on the performance
degradation of aperiodic tasks. For reducing overall energy consumptions,
aperioidic tasks are assigned to processors, capable of executing tasks at
lowest speed while ensuring system performance does not degrade beyond a
threshold level.
Design and Verification of Asynchronous Circuits and Systems:
Asynchronous circuits provide several alluring properties - over their
synchronous counterparts, such as locally generated timing signals in the
place of global clocks, potential performance speedups, robustness towards
variability in the manufacturing process and operating conditions, and
automatic shutdown of blocks that are not in use resulting in power savings.
The goals of the project are to develop:
1. Design of energy-efficient asynchronous circuits and systems.
2. Formal Verification methods for asynchronous circuits and systems that
are highly-automated, efficient, and scalable.
Complex relation discovery from distributed semantic web:
This project aims to discover complex yet meaningful and obscured
relationships between resource entities from the growing semantic web data.
For example, our project will enable users to uncover previously unknown and
potentially interesting correlations between two or more diseases, or
associations between different persons. The challenges include heterogeneous
data format, distributed data distribution, and large amount of data. We are
trying to combine the P2P semantic indexing and graph search to address this
problem.
Energy Efficient Green Radio: The ubiquitous wireless
applications and information technology increasingly contribute to the
overall energy consumption of the world. Reducing the energy requirement of
radio access networks has become an urgent environmental need. In this green
radio project, we consider both wide area networks (such as cellular) and
local area networks (such as WLAN). The objective is to investigate and
create innovative methods for the reduction of the total power needed to
operate a radio access network and to identify appropriate radio
architectures which enable such a power reduction. In particular, we focus
on the following two networks:
1. Future Mobile Cellular Radio Networks: We will develop new
energy efficient cellular deployment architectures by bringing BSs closer to
mobile users. However, increasing the density of BSs/RSs causes more
inter-cell interference (ICI). We will exploit cooperation between BSs/RSs
on how to turn interference into useful signals by the appropriate use of
radio resource management (RRM) techniques.
2. Future Enterprise and Home Networks: We will design and evaluate
adaptive resource management schemes in order to minimize energy consumption
while satisfying users’ QoS requirements. We will focus on adaptive
frequency assignment and network topology control for wireless mesh networks
and Femtocell networks to reduce energy consumption under dynamic traffic
loads. We will also consider dynamic spectrum access and resource management
to improve the coexistence and cooperation of heterogeneous wireless
networks.
Exploiting social collaboration for sharing over distributed
communities: In this project, we study how social network and
social collaborations can facilitate sharing over social communities. We
propose a novel scheme that exploits the social property of humans, such as
natural grouping, trust formation, and peer recommendation between people,
to facilitate sharing and collaborating in large-scale distributed
community. In this framework, network nodes perform local dynamic topology
adaptations to spontaneously create communities according to users’
social-closeness and trust. The basic premise of such semantic communities
is that sharing requests have a high probability of being fulfilled within
the community. Members in the same community share similar social patterns
hence are able to make recommendations to each other. We propose a fully
decentralized a trust-enabled recommendation system intended to overcome the
scalability barriers of centralized approaches. At the same time, social
trust is constructed between users according to their experiences and
recommendations from “friends”.
Semantics-based publish-subscribe for mobile ad-hoc networks
(MANET): MANETs are composed of autonomous, wireless devices which
are connected in a self-organized way without the need for fixed
infrastructure. These opportunistic and dynamic characteristics of MANETs
make communications in them complex. This publish/subscribe model decouples
time, space, and flow between publishers and subscribers, allowing for
greater scalability and a more dynamic network topology as required by
MANETs. However, the use of publish/subscribe paradigm in mobile ad hoc
networks brings many new challenges, mainly due to the highly dynamic
topology and the scarce resources. Most of the research on publish/subscribe
systems so far has concentrated on fixed network. Our project proposes an
effective semantics-based pub/sub communication scheme for large MANETs. The
main idea is to cluster users into zones. Efficient semantics-based intra-
and inter- zone communication is proposed to match subscriptions and
notifies the interested subscribers.
Cellular and Ad-hoc Hybrid Wireless Network: A hybrid
wireless network (HWN) consists of an infrastructure based network (such as
WiMAX) and several ad hoc components (such as MANETs). By using an HWN, one
can achieve the benefits offered by both infrastructure wireless networks
(e.g., good reliability and QoS support) and ad hoc networks (e.g., larger
coverage, low cost deployment, and flexibility). In this particular project,
we propose to investigate network design, Quality of Service provisioning,
and effective communications in HWNs. We will adopt a systematic,
joint-layer design approach for the proposed research. So far, we’ve modeled
the joint OFDMA resource allocation (in cellular) and QoS routing (in
ad-hoc) problem and proposed some good heuristic solutions.
Mobile social network: Mobile ad hoc social networks are
self-configuring social networks that connect users using mobile devices,
such as laptops, PDAs, and cellular phones. These social networks facilitate
users to form virtual communities of similar interests or commonalities.
This project aims to propose a concrete, generalized, and novel framework to
develop a fully functional mobile ad hoc social network. The proposed
framework provides effective and efficient solutions to social network
construction, semantics-based user profile matching, and multi-hop
semantics-based routing. Moreover, the proposed framework because of its
generality is applicable in applications of critical importance, such as
disaster-recovery, homeland security, and personnel control.
Efficient query evaluation over distributed data sites: The
broad availability of data coupled with increased capabilities and decreased
costs of both storage and computing technologies enable us to sharing data
in an unprecedented scale. This compels us to rethink how we will manage -
store, retrieve, explore, analyze, and communicate - this abundance of data.
In this project, we propose novel techniques that can greatly reduce
bandwidth cost of distributed query processing while allowing efficient
processing of complex queries, over large-scale, fully decentralized, and
semantically heterogeneous data sets. In particular, our solution integrates
topology adaptation, semantic query routing, and view-based caching
techniques to reduce bandwidth cost of distributed query processing while
allowing efficient evaluation of complex semantic queries over large-scale,
fully decentralized, and semantically heterogeneous networks.
Cellular and Ad-hoc Hybrid Wireless Network: A hybrid
wireless network (HWN) consists of an infrastructure based network (such as
WiMAX) and several ad hoc components (such as MANETs). By using an HWN, one
can achieve the benefits offered by both infrastructure wireless networks
(e.g., good reliability and QoS support) and ad hoc networks (e.g., larger
coverage, low cost deployment, and flexibility). In this particular project,
we propose to investigate network design, Quality of Service provisioning,
and effective communications in HWNs. We will adopt a systematic,
joint-layer design approach for the proposed research. So far, we’ve modeled
the joint OFDMA resource allocation (in cellular) and QoS routing (in
ad-hoc) problem and proposed some good heuristic solutions.
Distributed search/discovery: Locating desirable resources
and information from a large-scale distributed system such as P2P system and
grid is a very important issue. However, resource discovery in such systems
is challenging due to the considerable diversity, large number, dynamic
behavior, and geographical distribution of the resources. The resource
discovery technology required to achieve the ambitious global vision is
still in its infancy, and existing applications have difficulties in
achieving both rich searchability and good scalability. In this project, we
investigate the resource discovery problem for open-networked global-scale
networks. In particular, we propose a distributed semantics-based discovery
framework. We try to improve three aspects of performance: expressiveness,
scalability, and efficiency.