- Wireless Networks and Protocols
- Wireless Communication Networks Research
- Advanced Wireless Network Optimization
- IPv6, Mobility, Handover, Networks, Security
- Mobile Ad Hoc Networks
- Cooperative Communication and Network Coding
- Green IT and Sustainability
- Network Traffic and Congestion Control
- Caching and Content Delivery
- Advanced Neural Network Applications
- Advanced MIMO Systems Optimization
- Mobile Agent-Based Network Management
- Energy Harvesting in Wireless Networks
- Opportunistic and Delay-Tolerant Networks
- IoT and Edge/Fog Computing
- Topic Modeling
- Natural Language Processing Techniques
- Ferroelectric and Negative Capacitance Devices
- Advanced Data Storage Technologies
- Cloud Computing and Resource Management
- Machine Learning and Data Classification
- Distributed systems and fault tolerance
- Mobile Crowdsensing and Crowdsourcing
- Software System Performance and Reliability
- Software-Defined Networks and 5G
Microsoft Research (India)
2008-2025
Microsoft Research (United Kingdom)
2014-2024
Microsoft (United States)
2007-2020
Alcatel Lucent (Germany)
2000-2007
Nokia (United States)
1998-2007
Institute of Electrical and Electronics Engineers
2005-2006
Bell (Canada)
2005
Technion – Israel Institute of Technology
2003
University of Massachusetts Boston
1995-2002
AT&T (United States)
2002
We consider the problem of monitoring road and traffic conditions in a city. Prior work this area has required deployment dedicated sensors on vehicles and/or roadside, or tracking mobile phones by service providers. Furthermore, prior largely focused developed world, with its relatively simple flow patterns. In fact, cities developing regions, which comprise much tends to be more complex owing varied (e.g., potholed roads), chaotic lot braking honking), heterogeneous mix (2-wheelers,...
Mobile IP is the current standard for supporting macromobility of mobile hosts. However, in case micromobility support, there are several competing proposals. We present design, implementation and performance evaluation HAWAII (handoff-aware wireless access Internet infrastructure), a domain-based approach mobility. uses specialized path setup schemes which install host-based forwarding entries specific routers to support intra-domain micromobility. These deliver excellent by reducing...
Recent interest in supporting packet-audio applications over wide area networks has been fueled by the availability of low-cost, toll-quality workstation audio and demonstration that limited amounts interactive can be supported today's Internet. In such applications, received packets are buffered, their playout delayed at destination host order to compensate for variable network delays. The authors investigate performance four different algorithms adaptively adjusting delay an terminal...
In third-generation (3G) wireless data networks, mobile users experiencing poor channel quality usually have low data-rate connections with the base-station. Providing service to is required for maintaining fairness, but at cost of reducing cell's aggregate throughput. this paper, we propose Unified Cellular and Ad-Hoc Network (UCAN) architecture enhancing cell throughput, while fairness. UCAN, a client has both 3G cellular link IEEE 802.11-based peer-to-peer links. The base station forwards...
To realize the potential of opportunistic and participatory sensing using mobile smartphones, a key challenge is ensuring ease developing deploying such applications, without need for application writer to reinvent wheel each time. this end, we present Platform Remote Sensing Smartphones (PRISM) that balances interconnected goals generality, security, scalability. PRISM allows writers package their applications as executable binaries, which offers efficiency also flexibility reusing existing...
Cellular radios consume more power and suffer reduced data rate when the signal is weak. According to our measurements, communication energy per bit can be as much 6x higher weak than it strong. To realize savings, applications must preferentially communicate strong, either by deferring non-urgent or advancing anticipated coincide with periods of strong signal. Allowing perform such scheduling requires predicting strength, so that opportunities for energy-efficient anticipated. Furthermore,...
We introduce Gandiva, a new cluster scheduling framework that utilizes domain-specific knowledge to improve latency and efficiency of training deep learning models in GPU cluster.One key characteristic is feedback-driven exploration, where user often runs set jobs (or multi-job) achieve the best result for specific mission uses early feedback on accuracy dynamically prioritize or kill subset jobs; simultaneous entire multi-job critical. A second heterogeneity terms resource usage, making it...
As local area wireless networks based on the IEEE 802.11 standard see increasing public deployment, it is important to ensure that access network by different users remains fair. While fairness issues in have been studied before, this paper first focus TCP presence of both mobile senders and receivers. In paper, we evaluate extensively through analysis, simulation, experimentation interaction between MAC protocol TCP. We identify four regions unfairness depend buffer availability at base...
In 3G data networks, network operators would like to balance system throughput while serving users in a fair manner. This is achieved using the notion of proportional fairness. However, so far, fairness has been applied at each base station independently. Such an approach can result non-Pareto optimal bandwidth allocation when considering as whole. Therefore, it important consider network-wide context with user associations stations governed by optimizing generalized objective. this paper,...
Two important quality-of-service (QoS) measures for current cellular networks are the fractions of new and handoff "calls" that blocked due to unavailability "channels" (radio and/or computing resources). Based on these QoS measures, we derive optimal admission control policies three problems: minimizing a linear objective function call blocking probabilities (MINOBJ), probability with hard constraint (MINBLOCK) number channels constraints both (MINC). We show well-known guard channel policy...
We consider the problem of monitoring road and traffic conditions in a city. Prior work this area has required deployment dedicated sensors on vehicles and/or roadside, or tracking mobile phones by service providers. Furthermore, prior largely focused developed world, with its relatively simple flow patterns. In fact, cities developing regions, which comprise much tends to be more complex owing varied (e.g., potholed roads), chaotic lot braking honking), heterogeneous mix (2-wheelers,...
In many enterprises today, WAN optimizers are being deployed in order to eliminate redundancy network traffic and reduce access costs. this paper, we present the design implementation of EndRE, an alternate approach where elimination (RE) is provided as end system service. Unlike middleboxes, such benefits both end-to-end encrypted well on last-hop wireless links mobile devices.EndRE needs be fast, adaptive parsimonious memory usage opportunistically leverage resources hosts. Thus, a new...
Wireless link losses result in poor TCP throughput since are perceived as congestion by TCP, resulting source throttling. In order to mitigate this effect, 3G wireless designers have augmented their system with extensive local retransmission mechanisms. addition, increase throughput, intelligent channel state based scheduling also been introduced. While these mechanisms reduced the impact of on and improved utilization, gains come at expense increased delay rate variability. paper, we...
A large amount of popular content is transferred repeatedly across network links in the Internet. In recent years, protocol-independent redundancy elimination, which can remove duplicate strings from within arbitrary flows, has emerged as a powerful technique to improve efficiency face repeated data. Many vendors offer such elimination middleboxes effective bandwidth enterprise, data center and ISP alike.
Broadcast operation, which disseminates information network-wide, is very important in multi-hop wireless networks. Due to the broadcast nature of media, not all nodes need transmit order for message reach every node. Previous work on support can be classified as probabilistic (each node rebroadcasts a packet with given probability) or deterministic approaches (nodes pre-select few neighbors rebroadcasting). In this paper, we show how network-coding applied approaches, resulting significant...
WiFi radios in smart-phones consume a significant amount of power when active. The 802.11 standard allows these devices to save through an energy-conserving Power Save Mode (PSM). However, depending on the PSM implementation strategies used by clients/Access Points (APs), we find competing background traffic results one or more following negative consequences: increase, up 300%, client's energy consumption, decrease wireless network capacity due unnecessary retransmissions, and unfairness.
We present Gandivafair, a distributed, fair share scheduler that balances conflicting goals of efficiency and fairness in GPU clusters for deep learning training (DLT). Gandivafair provides performance isolation between users, enabling multiple users to single cluster, thus, maximizing cluster efficiency. is the first allocates cluster-wide time fairly among active users.
Systems for training massive deep learning models (billions of parameters) today assume and require specialized "hyperclusters": hundreds or thousands GPUs wired with high-bandwidth interconnects such as NV-Link Infiniband. Besides being expensive, dependence on hyperclusters custom high-speed inter-connects limits the size clusters, creating (a) scalability job parallelism; (b) resource fragmentation across hyperclusters.
In third-generation (3G) wireless data networks, mobile users experiencing poor channel quality usually have low data-rate connections with the base-station. Providing service to is required for maintaining fairness, but at cost of reducing cell's aggregate throughput. this paper, we propose Unified Cellular and Ad-Hoc Network (UCAN) architecture enhancing cell throughput, while fairness. UCAN, a client has both 3G cellular link IEEE 802.11-based peer-to-peer links. The base station forwards...
Next-generation wireless network standards are currently being defined. The access architectures have several specialized components tailored for their respective link technologies, even though the services provided by these different networks fairly similar. We propose a homogeneous IP-based as common technologies. uses Internet standard, Mobile IP, to support macro-mobility of mobile hosts, and HAWAII micro-mobility paging functionality current networks. also illustrate how proposed...
Mobile-IP is the current standard for supporting macro-mobility of mobile hosts. However, in case micro-mobility support, there are several competing proposals. In this paper we present design, implementation, and performance evaluation HAWAII: a domain-based approach mobility. HAWAII uses specialized path setup schemes which install host-based forwarding entries specific routers to support intra-domain micro-mobility. These deliver excellent by reducing mobility related disruption user...
In third-generation (3G) wireless data networks, providing service to low data-rate users is required for maintaining fairness, but at the cost of reducing cell's aggregate throughput. this paper, we propose unified cellular and ad hoc network (UCAN) architecture enhancing cell throughput while fairness. UCAN, a mobile client has both 3G interface IEEE 802.11 -based peer-to-peer links. The base station forwards packets destination clients with poor channel quality proxy better quality. then...