Kyung Wook Hwang

Kyung Wook Hwang

Ph.D. Candidate

Distributed Network Analysis (DNA) Group

Control, Management, and Telemedia (COMET) Group

Department of Electrical Engineering

Columbia University in the New York City

Office: CEPSR 807

Mailing address:
1300 Mudd
500 West 120th Street
New York, NY10027, USA

Phone: 212-854-7472

Email: kwhwang AT ee DOT columbia DOT edu

Education

Ph.D. Candidate in Electrical Engineering, Distributed Network Analysis (DNA) Group, Columbia University, New York, 2007~
Advisors: Prof. Dan Rubenstein and Prof. Vishal Misra

M.S. in Electrical Engineering, Columbia University, New York, 2007
GPA: 4.00/4.00

B.S. (First Class Honors) in Electronic Engineering, Sogang University, South Korea, 2005
GPA: 3.62/4.00

Visiting student in Electrical and Electronic Engineering, Sophia University, Japan, 1999~2000
GPA: 3.83/4.00

Research Interests

Distributed network algorithms

Content distribution and P2P networks

Network coding

Network security and privacy

Performance evaluation and modelling

Current Research Projects

Adaptive Bitrate Streaming in P2P VoD

I develop Joint-Family, a protocol that combines P2P and adaptive bitrate (ABR) streaming for VoD. While P2P and ABR have been proposed previously, they have rarely been studied together because they attempt to tackle different problems with seemingly orthogonal goals. I motivate my approach through analysis that overcomes a misconception resulting from prior analytical work, and explore that the popularity of a P2P swarm and seed staying time has a significant bearing on the achievable per-receiver download rate. I also show ABR in a P2P setting helps viewers achieve a much higher playback rate for the same number of interruptions. Peers in Joint-Family simultaneously participate in multiple swarms to exchange chunks of different bitrates. I adopt chunk, bitrate and peer selection policies that minimize the occurrence of interruptions while delivering high quality video while improving the efficiency of the system. Using traces, I compare Joint-Family to existing approaches for P2P VoD and show that viewers in Joint-Family enjoy higher playback rates with minimal interruption, irrespective of video popularity.

Abandonment and P2P VoD Streaming

The policies adopted in P2P VoD, however, have not taken user viewing behavior - that users abandon videos - into account. I show that abandonment can result in wasted resources. As a result, I reconsider the set of policies to use in the presence of abandonment. The goal of this project is to balance the conflicting needs of delivering videos without interruptions while minimizing wastage. I find that an Earliest-First chunk selection policy in conjunction with the Earliest-Deadline peer selection policy allows me to achieve high download rates. I take advantage of abandonment by converting peers to "partial seeds"; this increases capacity. I minimize wastage by using a playback lookahead window. I also do analysis and simulation experiments using real-world traces.

User Viewing Patterns for Optimal Video Placement

In VoD services I leveraged on the observation that users watch only a small portion of videos to optimize the placement of content. I investigated user viewing behavior from data collected from a nationally deployed Video-on-Demand service. I found that traditionally proposed prefix caches perform poorly unless there is a very large cache, because the working set of videos is large. Instead, I used a Mixed Integer Programming formulation to model the placement problem and employed an innovative solution approach that scaled well. My solution also took advantage of the ability to predict future demand of TV shows, while utilizing a relatively small cache to handle errors in demand estimation and flash crowds. I performed detailed simulations using traces of user viewing sessions (including stream control operations such as FastForward, and Rewind). My results showed that the use of segment-based placement strategy yielded substantial savings in both disk storage requirements at origin servers as well as network bandwidth.

Distributed Private Matching for On-Line Marketplaces

The Internet can serve as an on-line marketplace, enabling customers to select and utilize resources whose availability is advertised by providers. In many scenarios, however, both customers and providers want to ensure their privacy to the extent possible. I designed and developed a novel tool for ensuring privacy using secure multi-party computation (MPC). I then formulated a variety of marketplace scenarios in a way that was amenable to my tool, and apply my tool to achieve privacy in those scenarios. My tool easily handled marketplace exchanges involving tens of customers/providers and thousands of resources, and outperformed existing solutions based on secure MPC including Fairplay, VIFF, and SEPIA. This project illustrated that secure MPC could be a feasible approach for ensuring privacy in realistic network settings. My tool is available at http://www.ee.columbia.edu/~kwhwang/projects/gmw.html

Oblivious Shortest-Path Routing

I made a complete routing protocol where every node in the network determines the shortest path to a destination without any node learning any distance information except distance between the node and its neighbor nodes. Any node must not learn the distance to the destination. The motivation of this research was each node and the destination's location privacy. Existing shortest-path routing algorithms were not applied since they relied on distance information and found shortest paths based on that information. I used data encryption techniques and secure multiparty computation where a node can compute mathematical functions with its input and other nodes' inputs without sharing those inputs.

Anonymous and efficient multicast

Multicast applications often require security that is similar to what is offered for unicast. Anonymity is key to allowing nodes to hide their identity/location information from external observers, as well as from one another. We focus on providing anonymous multicast communications where messages are delivered from an anonymous source to multiple anonymous destinations. Our objective is to find ways to construct efficient multicast topologies in the traditional sense (short paths, minimal link utilization) without revealing information that is traditionally required to provide such efficient implementations. We show how shortest-path computations can be performed using an oblivious transfer mechanism so that nodes are able to route along the shortest path without learning the path length or the identities of the nodes along the path (other than their immediate neighbors).

Network coding in P2P network systems

In current peer-to-peer systems users often suffer from the slow down of the download speed in the beginning and the end of the download. When new users arrive, they do not have anything to offer to their neighbors and get low priority.(Tit-for-tat incentive mechanisms or choking policies may decrease speeds.) At the end of the download there is a last piece problem and the users cannot find the last missing blocks.(Local-rarest policies often fail to identify the globally rarest pieces when users have a limited view of the network.) To solve problems of existing P2P file distribution applications such as BitTorrent, network coding can be used. Each encoded codeword can be unique and useful to any user. Therefore, newly arriving users can easily obtain useful codewords which they can exchange with other users. Also at the end of the download users do not need to wait for a long time before they find their last missing pieces.

P2P streaming of stored media

Existing P2P networks is often not capable of allowing users to playback the stored media such as video and music files until the entire file has been downloaded. Systems such as Coolstreaming have presented live media content distribution using P2P network systems. However, stored media is different from live media where all participants are interested in the same part of the stream at the same time. For stored media users do not need to synchronize playing times. We explore whether P2P networking can be used to provide real-time playback capabilities for stored media. With real-time playback, users can start playback immediately after requesting the content. Also real-time playback should ensure uninterrupted playback of each user.

Professional Experience

Summer Intern, AT&T Labs Research, Florham Park, NJ, Summer 2012

Summer Intern, AT&T Labs Research, Florham Park, NJ, Summer 2011

Summer Intern, AT&T Labs Research, Florham Park, NJ, Summer 2010

Summer Intern, Bell Laboratories, Alcatel-Lucent, Murray Hill, NJ, Summer 2007

Programmer (Full time), Younglimwon Soft Lab Co., Ltd., Seoul, South Korea, 2003~2004

System Engineer and Programmer (Full time), NBIZ Technology Co., Ltd., Seoul, South Korea, 2001~2003

Teaching Experience

Teaching Assistant, Columbia University:

CSEE W4119 Computer Networks, Summer 2007

CSEE W4119 Computer Networks, Spring 2007

CSEE W4119 Computer Networks, Fall 2006

Certifications, Awards, and Scholarships

Extraordinary Teaching Assistant Award, Columbia University, Fall 2006

Certified Information Processing Engineer, Seoul, South Korea, 2000

Full Fellowship for Exchange Student Program, Tokyo, Japan, Apr.1999~Mar.2000

Japanese Language Proficiency Test (JLPT), Feb.1999

Academic Scholarship for Outstanding Grade, Sogang University

Last updated in Dec. 2012

Kyung Wook Hwang

Ph.D. Candidate

Control, Management, and Telemedia (COMET) Group

Department of Electrical Engineering

Office: CEPSR 807

Mailing address: 1300 Mudd 500 West 120th Street New York, NY10027, USA

Phone: 212-854-7472

Email: kwhwang AT ee DOT columbia DOT edu

Education

Research Interests

Current Research Projects

Professional Experience

Teaching Experience

Certifications, Awards, and Scholarships

Mailing address:
1300 Mudd
500 West 120th Street
New York, NY10027, USA