Research
A.
Cloud Computing
The
emergence of user-facing online services such as web search (e.g., Bing), social
networks (e.g., Twitter), and advertising systems (e.g., Google AdWords) has
led to the development of mega-scale datacenters. In addition, with the rise of
cloud computing service providers like Amazon, Microsoft, and Google,
datacenters have emerged as a primary platform for managing the computation,
storage, and communication requirements of users. These datacenters have unique
network and traffic characteristics (e.g., microsecond latencies,
partition-aggregate structure of applications, and multi-rooted tree
topologies) that give rise to interesting research questions such as: what
topologies work best for datacenters? how can we design efficient load
balancing schemes? how can we architect scalable cloud management frameworks?
how can we come up with power efficient datacenter designs suited to the needs
of the developing world?
[DAS, CoNEXT 2019]
[2D, CoNEXT 2018]
[DRIBS, NSDI 2017p]
[SAPS, INFOCOM 2017]
[RANS, HotNets 2016]
[eSDN, SIGCOMM 2015p]
[MulBuff, SIGCOMM 2014p]
[PASE, SIGCOMM 2014] (PASE ns2
simulation code,
patch applies to ns-2.34)
[L2DCT, INFOCOM 2013] (L2DCT ns2 simulation code, patch applies to ns-2.35)
[Coexistence of Datacenter Transports,
ICC 2014]
[Incast Congestion, Globecom 2015]
[RACS, ICC 2013]
B.
ICT for Development + Countering
Misinformation
Despite the Internet's massive success and our
increasing reliance on it, the fruits of its success are not distributed
equally. For example, access to affordable Internet in developing countries
remains low, and Internet is at least 10x slower in (least) developing
countries compared to developed countries, Thus, the common use of low-end mobile
devices along with a slow Internet connection often leads to poor user
experience in developing countries. Coupled with lack of digital literacy and
skills, we see large digital divides between
developed and developing countries. Many factors have contributed to this
situation including lower per-capita income levels, poor Internet
infrastructure, and lack of basic ICT skills. Our research aims to bridge these
digital divides.
[Audio Deepfake Detection Using Attention DNNs,
ASRU, 2021]
[Affordable Web Architecture, HotNets 2021]
[Robust Detection of Audio Deepfakes,
INTERSPEECH 2021]
[Mobile Web Browsing Under Memory Pressure,
SIGCOMM CCR 2020]
[Google Web Light, WWW 2020]
[MissIt, CHI 2020]
[Educational Interventions for Combating Fake News,
AEA 2020]
[Mobile Video, SIGCOMM 2019p,
IMC 2018p]
[Mobile Web, IMC 2018p]
[Facebook Free Basics, SIGCOMM CCR 2017
– Best Paper Award]
[Facebook Free Basics, SIGMETRICS 2017p]
[Mobile Phone Characteristics in Developing Regions,
IMC 2016]
[Long Distance WiFi, INFOCOM 2014p]
C.
Internet Censorship
Internet
censorship has become increasingly pervasive with nearly 70 countries
restricting internet access in one way or another. The resulting impact on the
user base is widespread and has drawn a lot of interest from researchers,
activists, and citizens in recent years. Existing systems research on Internet
censorship has focused on either measuring censorship (i.e., what is blocked,
where it is blocked, how it is blocked, and when it is blocked?) or designing
tools for resisting censorship (e.g., Tor). We are exploring several research
themes in this space: How can we design scalable systems for gathering reliable
and continuous censorship measurements? How can one improve the performance of
anonymity tools? How can we analyze mechanism(s) used by a censor to block
content in real-time? What are the ethical and effectiveness tradeoffs in
measuring censorship?
[Censorship Economics, AEA 2020]
[C-Saw, SIGCOMM 2018, HotNets 2015]
[Ethics, NSDI 2018p]
[Advention, HotNets 2017]
D.
Mobile/Wireless
Networks
Wireless
technologies (e.g., WiFi, Bluetooth, and 3G/4G) have become extremely pervasive
in our lives. While technological advances have led to substantial improvements
in these technologies, several research challenges arise as these technologies
increase in diversity, scale (e.g., in terms of speeds), or are densely
deployed. These trends lead to many interesting challenges such as (1) how can
we use multiple wireless interfaces (e.g., WiFi, Bluetooth, and 4G) to better
support different applications (e.g., AR/VR), (2) how can we achieve high
throughput efficiency at Gigabits/sec WiFi speeds?, (3) how can we design a
scalable packet core for next-generation cellular networks (e.g., 5G and
beyond), and (4) how can we design extremely low cost WiFi that can achieve
high speeds at 10s of kilometers of distances?
[Scylla, CoNEXT 2018]
[SlickFi, CoNEXT 2016]
[BLMon, INFOCOM 2014, INFOCOM 2013]
[Rate Equilibria in WLANs, LCN 2013]
E.
Greening the
Electrical Grid using Internet Concepts
The Electrical
grid and the Internet share the same goal of interconnecting distributed
suppliers to geographically dispersed consumers. The existing grid faces
several challenges: it is centralized, with wasteful
transmission and distribution technologies, carbon-intensive energy sources,
limited energy storage, and minimal use of ICT. We think the future Electrical
Grid will have architecture like the Internet and thus the lessons learnt from
its design and evolution could play a key role towards the Next-Generation Grid
(a.k.a Smart Grid).
[Models for Load Prediction, ICONIP 2012]