Binh Nguyen's CV

Education

• Ph.D in Computer Science, University of Utah, 2017 (GPA 3.9/4.0)
  • Thesis: “Enhancing scalability and reliablility in mobile core networks”.
  • Advisor: Jacobus Van der Merwe.
• B.S. in Computer Science, Shanghai Jiao Tong University, 2012 (GPA 3.7/4.0)

Work experience

• Charter Communications - Principal Engineer - Denver - 2017/12 - Now
• Microsoft Research - Research Intern - Cambridge, UK - Summer 2016
• Nokia Bell Labs - Research Intern - Murray Hill, NJ - Summer 2015
• AT&T Labs Research - Research Intern - Bedminster, NJ - Summer 2014
• Flux Research Group - Research Assistant - Univeristy of Utah - 2012-2017
• School of Computing - Teaching Assistant - Univeristy of Utah - Fall 2012

Skills

• LTE/mobile network: LTE stack (RLC, PDCP, GTP-C/U, S1AP, S11, S5/S8), Software Defined Radio (OpenAirInterface, IP.Access), EPC Core (OpenEPC).
• Data analytics: Hadoop, Pig, ELK stack, Kafka, Avro, Anomaly detections.
• Infrastructure: Kubernetes, Docker containers.
• Programming Languages: Python, C, Bash.
• Network/automation: SDN controllers (Ryu controller, Open Daylight), Open vSwitch, ONOS, YANG.
• Network protocols: TCP, OSPF, Segment Routing, SNMP.
• Other tools: NS3, Mininet, Emulab, PhantomNet, Free Range Routing.

Publications:

ENHANCING SCALABILITY AND RELIABILITY IN MOBILE CORE NETWORKS.
Binh Nguyen, Ph.D. Dissertation. University of Utah. 2018.
[Ph.D. Dissertation].

ECHO: A reliable distributed cellular core network for hyper-scale public clouds.
Binh Nguyen, Tian Zhang, Bozidar Radunovic, Ryan Stutsman, Thomas Karagiannis, Jakub Kocur, and Jacobus (Kobus) Van der Merwe. MOBICOM’ 18. New Delhi. India.
Paper 1-min video.

SIMECA: SDN-based IoT Mobile Edge Cloud Architecture.
Binh Nguyen, Nakjung Choi, Marina Thottan, and Jacobus Van der Merwe. IFIP/IEEE International Symposium on Intergrated Network Management 2017 (ISINM’ 17), Portugal.
Paper Repeatable profile SDN controller source code

ABSENCE: Usage-based Failure Detection in Mobile Networks..
Binh Nguyen, Zihui Ge, Jacobus Van der Merwe, He Yan, and Jennifer Yates. MOBICOM’ 15. Paris, France.
Paper 1-min video Slides

PhantomNet: Research Infrastructure for Mobile Networking, Cloud Computing and Software-Defined Networking.
Arijit Banerjee, Junguk Cho, Eric Eide, Jonathon Duerig, Binh Nguyen, Robert Ricci, Jacobus (Kobus) Van der Merwe, Kirk Webb, and Gary Wong. ACM GetMobile’ 16.
Paper

Efficient, Adaptive and Scalable Device Activation for M2M Communications.
Arijit Banerjee, Binh Nguyen, Vijay Gopalakrishnan, Sneha Kumar Kasera, Seungjoon Lee, and Jacobus Van der Merwe. IEEE SECON’ 15.
Paper

Towards Understanding TCP Performance on LTE/EPC Mobile Networks.
Binh Nguyen, Arijit Banerjee, Vijay Gopalakrishnan, Sneha Kumar Kasera, Seungjoon Lee, Aman Shaikh, and Jacobus Van der Merwe. ACM SIGCOMM All Things Cellular’ 14, Chicago.
Paper Slides

SMORE: Software-Defined Networking Mobile Offloading Architecture.
Junguk Cho, Binh Nguyen, Arijit Banerjee, Robert Ricci, Jacobus Van der Merwe, and Kirk Webb. ACM SIGCOMM All Things Cellular’ 14, Chicago.
Paper Slides

Patents:

Telecommunications network with data centre deployment.
Bozidar Radunovic, Christos Gkantsidis, Thomas Karagiannis, Parisa Jalili Marandi, Binh Nguyen, Matthew John Balkwill.
US, 15406348. 7/19/2018.

Programmable system architecture for routing data packets in virtual base stations.
Nakjung Choi, Binh Nguyen, Marina Thottan.
US, 15068953, 9/14/2017.

Awards:

• NSF travel grant for Mobicom 2015, Paris, France.
• Scholarship to studying aboard for excellent students granted by Vietnam Ministry of Education in 2007.
• Excellent undergraduate student scholarship by Shanghai Municipal Government in 2009 & 2010 & 2011.
• Fist Prize High School Physics competition (region level) in 2005 & 2006.

Teaching

• Teaching assistant, University of Utah, Fall 2012, CS3810 Computer Architecture (Undergrad)

Service

• Reviewer for 2014 IEEE/ACM Transactions On Networking (ToN).
• Volunteer for 2017 ACM Mobicom conference, Snowbird, UT.

Projects:

• PhantomNet - end-to-end LTE/EPC testbed (Spring 2012 - Now)
  I worked on helping to deliver functionalities in PhantomNet such as: running and testing Software Defined Radio (SDR) eNodeB, work with OpenEPC’s source code to extend EPC protocols (e.g., S1 handover) in OpenEPC implementation, automating the provisioning of subscriber information (e.g., populate HSS database with subscriber information) in EPC core network. I also created PhantomNet profiles as packaged environments of end-to-end LTE/EPC mobile networks. The profiled LTE/EPC networks include different combinations of: Nexus 5/simulated SDR UE/emulated UE, SDR eNodeB/emulated eNodeB, and SDR-EPC/OpenEPC core networks. I created tutorials for the users of PhantomNet that explain the profiles and how to use them.

• Large scale failure detection in mobile networks (Spring 2015)
  Despite the existing monitoring systems, there are still failures that happen without awareness of the network operator. Those are silent failures as the operator only learn about them when customers start to complain. Unfortunately, detecting all service disruptions is challenging. Somewhat counterintuitively, monitoring network components is insufficient to indicate user experience because of a complex relationship between the network status and user-perceived service experience. Moreover, service disruptions could happen because of reasons that are beyond the network itself, e.g., a bug in a firmware update causing all devices of a model not being able to connect. We proposed ABSENCE, a passive service monitoring system that detects service disruptions by monitoring customer usage. The main idea is that customer usage of a large-enough aggregation of users is predictable. A network problem eventually manifests as a drop in customer usage. Therefore, usage of an aggregation of customers, or lack thereof, is a reliable indicator of service disruptions or network problems. I designed and built ABSENCE. It uses users’ Call Data Records (CDR) to obtains usage of aggregations of customers across multiple dimensions (e.g., ZIP code, device made, model, service types). From the usage time series, ABSENCE uses the time decomposition technique to predict future usage and infers an anomaly if the usage is less than expected. ABSENCE uses Map-Reduce to aggregate users in parallel from Tera-Bytes of CDR generated constantly in the network. ABSENCE was able to detect failures with up to 95% of accuracy within the first hour of outage. It was deployed on a Hadoop cluster with 200 nodes in the operator and detected real failures that went under the radar.

• Novel SDN-based mobile edge cloud architecture for a massive number of Internet-ofThings devices (Spring 2016)
  By 2020, there are 26 billions of IoT devices (source: Gartner). A large portion of them will use the mobile network because of the large coverage and the better security. However, the current EPC core network was designed to support a limited number of devices streaming high-quality traffic (e.g., smart phones streaming a Youtube video), but not a large amount of devices sending small and sporadic traffic (e.g., billions of IoT devices sending a few bytes every 15 minutes). I proposed SIMECA, a SDN-based mobile edge cloud architecture that provides a new network service abstraction that is more suitable for IoT devices. Unlike LTE/EPC data plane that uses GPRS tunnels (GTP) and maintains per device state in the core network (i.e., SGW, PGW), SIMECA’s data plane only maintain per device state at the base station. This helps reduce SIMECA’s control plane overhead because connection set up only requires installing classification rules at the base station. To reduce packet header overhead, SIMECA translate packet header at base station without adding extra GTP header into it. The header translation mechanism also allows SIMECA to separate device identity and routing identity and support seamless mobility. I implemented SIMECA by extending OpenAirInterface’s SDR eNodeB’s data plane and OpenEPC’s MME implementation. I evaluated SIMECA using real SDR hardware and smart phones. Compared to LTE/EPC, SIMECA’s control plane and data planes are 20% and 37% less overhead respectively.

• Enhancing availability of NFV-based mobile networks in public clouds (Spring 2017)
  With technology advancements in SDN and NFV, next generation mobile networks are expected to run on Commodity Off-the-shelf (OTS) hardware. However, in contrast to telecom-grade hardware with built-in redundancy (five 9s availability), commodity Off-the-Shell (COTS) hardware wasn’t built with reliability (often guarantee three 9s availability). We proposed ECHO, a mobile core architecture that uses existing distributed systems techniques to improve availability of the software-based mobile core network running on an unreliable infrastructure. ECHO is designed to scale, to be highly available despite failures while behaving exactly like the EPC core network that runs on reliable hardware. ECHO is general as its operation does not depend on the correctness of the LTE/EPC protocols that also constantly change. ECHO proposes to decouple state from computation in the mobile core components; stateless redundant computation instances and replicated state storage. ECHO leverage a necessarily reliable agent at each base station that keeps sending a request until it receives a reply. Under duplication of requests and failures, ECHO needs to (i) assure atomic state change on each EPC component and across components, and (ii) maintain in-order execution of requests issued by the mobile device. To achieve atomic state change, ECHO uses a non-blocking algorithm with Optimistic Concurrency Control to ensure only one stateless instance can make progress. To achieve in-order execution, ECHO tags a sequence ID to each request at the agent at the base station and ensures ECHO’s EPC components execute the requests in the order that is specified by the sequence ID. I implemented ECHO and evaluated it on PhantomNet and Microsoft Azure with real LTE small cells and a mobile smart-phones. ECHO can achieve five 9s availability with 10% of overhead.

• *Offloading architecture in mobile network using Software-defined network (Fall 2014)
  4G mobile core networks are largely based on proprietary and expensive hardware components that limit both new functionalities as well as low latency applications. SMORE is an SDN-based offloading architecture for LTE/EPC network. Using an SDN infrastructure and an SDN controller, together with a proper monitoring mechanism, SMORE offloads a portion of LTE traffic to a local edge cloud to support low latency applications. The challenges of SMORE are to find a practical solution given the current deployment of the mobile network and to be transparent to existing LTE traffic. SMORE intercepts data and control plane traffic at Mobile Telephone Switching Office (MTSO). It monitors the LTE control channel to extract data plane information used in data plane offloading. It offloads registered traffic to the mobile edge cloud and leaves normal LTE traffic untouched. I implemented SMORE’s SDN controller (using Ryu) and SMORE’s monitor (using tshark).

• Understanding TCP performance in LTE/EPC mobile networks (Fall 2014)
  Because of the closed nature of mobile networks, many studies treat the network as a black-box and try to probe the network from the outside to learn insights about it. In this project we wanted to look under the hood to understand how LTE/EPC’s network protocols and procedures could potentially cause problems with TCP performance. By extending the LTE module in the NS3 simulator, I conducted experiments to compare TCP performance in two types of handover: lossless handover and seamless handover. I also conducted experiments to study the impact of sudden load increase on a cell on TCP performance. My first finding is that lossless handover imposes no packet loss during a handover however incurs high end-to-end delay because of queue forwarding, while seamless handover causes many packet loss but does not increase the end-to-end delay because the forwarding queue is discarded during a handover. The second finding is that a sudden load increase at a base station could cause a sudden bandwidth throttle of other devices in that base station. That sudden bandwidth throttle causes a sudden delay peak at the TCP level and TCP spurious timeouts. Those TCP spurious timeouts could cause the TCP congestion window to drop to near zero and affect TCP performance. The findings suggest cross-layer interaction between end-devices and the mobile network to maximize applications’ performance.