Daniel A. Freedman
Post-Doctoral
Research
Associate
Department of
Computer Science
Cornell University
| email: | dfreedman@[see-ess cornell edu] |
| work: | 607-255-[zeero won won zeero] |
| office: |
Cornell University Department of Computer Science 4119C Upson Hall Ithaca, NY 14853 |
Research
My research agenda spans distributed systems and networking.
As a reflection of my academic upbringing, I draw inspiration from physics to address questions in computer science and engineering.
In this vein, my recent work pushes the state-of-the-art in precision and reproducibility of network measurements, which ultimately reveals insights into distributed-systems behavior, informs architecture design, and leads to more dependable systems.
I also pursue separate, more traditional, distributed-systems projects, building systems for wide-area collaboration that integrate server-hosted data with edge content, and that optimize wide-area communication via multicast overlays on complex heterogeneous networks.
Academic background
In Fall, 2008, I joined the Computer Science Department at Cornell University as a Post-Doctoral Research Associate, where I work with Ken Birman on (distributed) systems and networking.
Prior, I received my M.S. and Ph.D. in theoretical physics, also from Cornell, and my S.B. in physics from MIT. Professor Tomás Arias served as my advisor for the former graduate degrees, and Prof. Dr. Uwe-Jens Wiese (now at the Albert Einstein Center for Fundamental Physics at the University of Bern) for the latter undergraduate one.
My doctoral work in condensed-matter physics involved computational studies, both empirical and ab initio, of the growth of crystalline oxides.
Advising
As a post-doc (and a graduate student before), I've supervised a significant number of both M.Eng. students and undergraduates:
Computer Science (M.Eng.): Jared Cantwell (2008-10), Petko Nikolov (2008-9), Yilin Qin (2009), Mihir Patel (2009-10), Michael Ryan (2010), Matt Pearson (2010), Matthew Mukerjee (2010-11), Samarth Lad (2011)
Computer Science (U.Grad.): Chuck Sakoda (2008-9), Revant Kapoor (2009)
Computer Science (H.S. [Outreach]): Theo Jepsen (2009), Nadia Kiamilev (2010), Jed Thompson (2011)
Electrical and Computer Engineering (M.Eng.): Hera Li (2010-11)
Electrical Engineering (Visiting U.Grad.): Sushobhan Nayak (2010, IIT-Kanpur)
Physics (U.Grad.): Anselm Levskaya (2002-3), Carlos Ramirez (2006-8)
I welcome students to contact me regarding research possibilities. I have opportunities not only for folks from Computer Science, but also from Cornell's Electrical and Computer Engineering and Applied and Engineering Physics departments.
Spring get-together in April, 2011, with current students in then-research group: Hera Li, Samarth Lad, Daniel Freedman, Kyle Hsu (Kyle left shortly thereafter, without participating in group research), Matt Mukerjee (L to R). (Earlier group photo from December, 2010.)
Current projects
Instrumentation for exact network
timings:
I invented and implemented BiFocals
instrumentation to provide the first-ever exact measurement of
timings of network packets in flight. My team deployed
BiFocals to empirically characterize 10 Gbps Ethernet (10GbE)
optical wide-area networks (see below), both in the wild and
in the lab.
BiFocals substantially departs from typical network-timing
techniques that use measurement software running on endpoint
computers attached to the network under test; these invariably
introduce non-deterministic timing errors. BiFocals instead
directly taps the network fiber. It leverages common physics
test equipment — oscilloscopes, pattern generators,
frequency synthesizers, lasers, etc. — as well as
numerical methods to deliver provably exact timings. It
records a contiguous portion of the entire physical-layer
symbol stream in real-time. Software post-processing utilizes
Fast Fourier Transforms for clock recovery, protocol knowledge
for packet extraction, and symbol-stream timebase information
to assign the final time-stamps. Our precision is
significantly better than the width of a single 10GbE symbol
(~100 picoseconds), hence our timings are exact, even at the
highest deployed data rates.
The BiFocals software components are released under an
open-source BSD license and available through http://bifocals.cs.cornell.edu.
High-speed WAN measurement and protocol
design:
Using the precision afforded by BiFocals, I discovered a novel
class of packet burstiness on wide-area network (WAN) paths.
(A public
dataset of the captured traces is available.) We
identified the formation of packet chains as traffic transits
WANs, even with the network amply over-provisioned. We
measured endpoint packet loss, induced by such burstiness, and
found it to be far more severe than expected from packet-chain
lengths. We recognized this as deriving from an underlying
impedance mismatch between ever-increasing raw network data
rates and constrained endpoint single-core processing power.
We are now designing new end-to-end parallel-stream
data-transfer protocols to prevent such endpoint loss upon
exposure to packet chains and to harness modern commodity
multi-core architectures.
Systems for wide-area collaboration:
Kevlar optimizes group-based communication across
qualitatively different network environments, by federating
multiple application- and physical-layer multicast protocols.
In doing so, it provides a feasible and deployable
“narrow waist” for large-scale multicast. It
delivers the first practical multicast across realistic WAN
deployments that combine backbone transit segments (generally
without any physical-layer multicast) with extensive local
topologies, themselves supporting various flavors of multicast
and burdened with firewalls, bottleneck links, and other local
network-policy impediments. Due to its decentralized
membership state, Kevlar provides resilience to churn and
efficiency of recovery, critical for real deployments in
non-trivial applications and settings. Kevlar builds atop our
Live Objects platform for distributed programming.
Our Live Objects platform allows novice programmers to create
complex distributed systems, just as conceptually simple
markup languages lowered barriers to create web sites and
applications. Live Objects extends the ideas of
language-based composition to include communication channels:
It replicates objects on multiple nodes in a distributed
system, synchronizes state across the network, and forms
compositions on each node that interact via event-based
interfaces. This facilitates combination of server-hosted data
with edge content, using both peer-to-peer and client-server
technologies.
Selected publications
Consistent Soft-State Replication in the Cloud: How Amnesia-Freedom in Isis2 Evades the CAP Theorem.
Kenneth Birman, Patrick Dowell, Daniel Freedman, Qi Huang.
In submission.
Instrumentation for exact packet timings in networks.
Daniel A. Freedman, Tudor Marian, Jennifer H. Lee, Ken Birman, Hakim Weatherspoon, Chris Xu.
Exact temporal characterization of 10 Gbps optical wide-area network.
Daniel A. Freedman, Tudor Marian, Jennifer H. Lee, Ken Birman, Hakim Weatherspoon, Chris Xu.
(Early-Accept Paper; one of three.)
Empirical Characterization of Uncongested Lambda Networks and 10GbE Commodity Endpoints.
Tudor Marian, Daniel A. Freedman, Hakim Weatherspoon, Ken Birman.
Kevlar: A Flexible Infrastructure for Wide-area Collaborative Applications.
Qi Huang, Daniel A. Freedman, Ymir Vigfusson, Ken Birman, Bo Peng.
Enabling Tactical Edge Mashups with Live Objects.
Daniel Freedman, Ken Birman, Krzys Ostrowski, Mark Linderman, Robert Hillman, Albert Frantz
(Best Paper in Track; Best Paper in Conference.)
Daniel A. Freedman, D. Roundy, T. A. Arias.
Physical Review B 80, 064108 (2009).
(Selected by American Physical Society journal editors for inclusion in Physical Review B's Kaleidoscope Images for August, 2009.)
Support
The National Defense Science and Engineering Graduate (NDSEG) Fellowship, awarded by the United States Department of Defense, provided a measure of independence for the start of my doctoral program.
My research funding continued through the Cornell Center for Materials Research (CCMR) and, specifically, its Interdisciplinary Research Group (IRG) on the Dynamics of Growth of Complex Materials.
Within the Computer Science department, much of my recent funding is provided by the Directorate for Computer & Information Science & Engineering (CISE) of the National Science Foundation (NSF), the Information Directorate of the Air Force Research Laboratory (AFRL), and the Air Force Office of Warfighting Integration (AF/XC).
Personal
I am a passionate (yet quite) amateur photographer (with focus on portraiture and macro work), Free Software proponent, and horology enthusiast...