|
Graduate Research Assistant at
Real Time and Distributed Systems Laboratory (RADS Labs), Department of
Systems and Computer Engineering, Carleton University under the
supervision of Prof. Shikharesh Majumdar and Dr. Eric Parsons (Nortel
Networks).
Major Projects:
-
A Framework for QoS Aware Resource Management in Multi-Organizational
Grids (in collaboration with Nortel Networks): Since
there are multiple stakeholders in Grids – resource consumers and
resource owners – each with different objectives and since Grid
resources are shared by multiple applications which may belong to
different administrative domains, it is difficult to meet QoS objectives
of applications while maintaining high system performance that is
important to resource owners. I presented a complete framework for
resource sharing in multi-institutional Grids that can meet the
objectives of the both stakeholders in Grids. The framework relies on
the notion of under-constrained advance reservation requests which have
laxity in their reservation windows. The framework provides components
for each of the fabric, resource and collective layer of the Grid
architecture and aligns well with other Grid technologies. Several
components of the framework are being integrated into the well-known
Globus Toolkit and will be available as an add-on patch for users around
the globe.
-
Dynamic Scheduling of Lightpaths in Lambda Grids (in
collaboration with Nortel Networks): Dynamic optical networks hold
the potential of satisfying very large bandwidth requirements of many of
the Grid applications. However, encapsulation of optical network
elements into manageable Grid resources and dynamic provisioning of
lightpaths is necessary to meet the complex demand patterns of the Grid
applications. In this project, I presented a scalable algorithm (Scaling
through Subset Scheduling Algorithm) for an NP-Hard problem of
scheduling on-demand and advance reservation requests for lightpaths.
The algorithm meets the demands of applications while seeking to
optimize usage of optical network. I also investigated the impact of
data segmentation on performance and demonstrated how laxity in the
reservation window can be exchanged for segmentation to achieve high
utilization of lightpaths.
-
Dynamic Matchmaking in Multi-Institutional Environments (in
collaboration with Nortel Networks): In any form of distributed
computing effective application-to-resource mapping and load balancing
is important to meet performance objectives. This process is generally
referred to as matchmaking. In this project, I demonstrated that
traditional matchmaking algorithms used in distributed computing result
in poor performance when applied in multi-institutional settings
particularly for workloads consisting of both advance reservations and
best effort jobs. I introduced a novel algorithm for matchmaking,
Minimum Laxity Impact (MLI) that outperforms all other algorithms
investigated in almost every respect for a wide range of workload
parameters. In comparison to other algorithms, MLI results in the lowest
response time for the users while providing the highest resource
utilizations. Another application of MLI is as an effective
meta-scheduler within Platform’s Community Scheduler Framework (CSF)
that is used to dispatch jobs to resource managers in multiple domains.
MLI is currently being integrated into CSF.
-
Configurable Grid Middleware: Modern resource sharing systems
comprise thousands of resources and consumers. On such a scale, dynamic
configuration of the system in accordance with workload conditions is
necessary for high efficiency. In this project, I introduced
configurable Grid entities that adapt to meet the requirements of the
system. I presented a novel heuristic-based scalable algorithm, Grid
Scheduling with Deadlines (GSD), for an NP-Complete problem of
scheduling advance reservation and on-demand requests on shared Grid
resources where a number of Grid jobs can run in parallel with each
other. GSD supports non-preemptable and preemptable jobs and adapts
itself to the workload conditions. I also introduced components that
adapt the resource schedule to prevent performance degradation resulting
from exceptions due to inaccurately estimated runtimes and abnormal
terminations of jobs.
-
Engineering Mobile Wireless Publish/Subscribe Systems for High
Performance: Decoupling and asynchronous nature of publish/subscribe
paradigm makes it a good choice for mobile wireless networks domain. In
this project, I showed that still the current implementations of
publish/subscribe systems as well as the traditional solutions for
extending publish/subscribe systems to the mobile domain do not perform
well in highly mobile and unreliable wireless settings. I introduced a
novel approach called semi-durable subscriptions for extending
publish/subscribe systems to mobile wireless networks. Semi-durable
subscriptions are computationally inexpensive and they rely on efficient
caching of information and novel cache replacement policies to prevent
message loss in unreliable and dynamic mobile wireless settings.
Experimental investigation has attested to its efficacy and superiority
over traditional solutions.
-
DEVS Modeling of Mobile Wireless Ad Hoc Networks: Analyzing
wireless ad hoc networks is a complex task due to their dynamic and
irregular nature. Cellular Automata (CA), a very popular technique to
study self-organizing systems, can be used to model and simulate ad hoc
networks, as the modeling technique resembles the system being modeled.
Cell-DEVS was proposed as an extension to CA in which each cell in the
system is considered as a DEVS model. In this project we show how these
techniques can be used to model mobile wireless ad hoc networks, making
easy model definition, analysis and visualization of the results. With
the help of Cell-DEVS we were able to extend ad hoc routing techniques
for internetworking and multicast routing.
-
Context-Aware Mobile Computing: There has been an ever-increasing
demand from consumers to be able to access services accessible from
desktops through devices such as cell phones and portable digital
assistants without any interruptions. However, user mobility and
unreliable wireless channels make it challenging for the developers to
provide services seamlessly to the mobile user. My research explored how
mobile wireless devices can be seamlessly integrated with wired networks
and how we can exploit context awareness to provide a unique set of
services targeted specifically for these environments. I designed a
thin-client architecture for accessing location-dependent services over
short-range communications.
-
High Performance Software Systems: Evaluating performance of
software systems early in their life cycle helps in determining software
bottlenecks and enables redesign (if necessary) without incurring
substantial time and monetary costs. The process becomes even more
important for multi-tier distributed applications where a bottleneck in
one of the tiers can have devastating consequences for the whole system.
I did scalability analysis of multi-tier distributed software patterns
using layered queuing networks.
The
research provides an insight into the scalability behavior of different
patterns. My results show that the choice of a scaling strategy should
not only take into account the pattern being scaled up but also factors
such as the scale factor to which the scalability is desired and the
size of the pattern
|
