The Edgio Labs and Advanced Projects team is a group of systems and networking researchers who explore new ways in which we can improve the performance, reliability, and capabilities of the Edgio platform. In this article, we’ll give an overview of the team, their backgrounds, and some of our primary areas of focus, including some of our past work. We’ll also discuss how we approach problems in a way that enables effective and pragmatic design. In upcoming articles, we’ll provide a deep dive into some of our ongoing work, but, for now, let’s start things off with an overview.
Who We Are
Kyriakos Zarifis is a Research Scientist who joined the team in 2019. Kyriakos received his PhD from the University of Southern California in 2018, during which he focused largely on HTTP optimizations for Content Delivery Networks (CDNs). Currently, his research interests include Internet measurements and Web protocol optimization across the Edgio network, transport, and application layer.
Leading the team is Marcel Flores. Marcel was originally an intern in the summer of 2014 and joined the team full-time after completing his PhD at Northwestern University, with a focus on enabling additional channels of communication in existing networks and improving network performance. At Edgio he has further explored transport layer optimizations, large-scale traffic management strategies, and cache optimizations.
Anant Shah, Principal Architect, is a frequent contributor to the team and its projects. Anant’s research interests include large-scale Internet measurements, routing analysis, and traffic optimizations. His current focus is on optimizing and scaling backend API services and automated deployment analysis. Anant received his PhD from Colorado State University in 2018.
Our Areas of Focus
Over the years, in studying the needs of the infrastructure, the team has found that there are a few areas of focus that drive the majority of our projects.
First among these, is the general study of Internet measurement and understanding how the Edgio platform interacts with the wider internet. This includes understanding the relationships with different provider networks, and how they can be best utilized to offer the lowest latency and highest reliability. For example, past projects have explored how to optimize anycast announcements through a sequence of controlled experiments, how to best measure common points of failure in our client-facing routes, or how to take advantage of public Border Gateway Protocol (BGP) announcements to offer early warning of disruptions to our infrastructure and the infrastructure we are dependent on. Other work has more directly considered how to measure application-level behaviors, exploring questions on how we can estimate users’ video QoE experience.
Another area of focus has been on the need to optimize existing systems on the infrastructure, understanding how we can improve the performance of the millions of daily transactions serviced by the network. This type of work generally arises from the examination of performance bottlenecks in the day-to-day operations of the network. These projects cover a number of stack components: helping to understand how data is flowing through our physical infrastructure, and what types of congestion control provide the best performance in different conditions. At the application level, we’ve worked to optimize systems to enable more efficient autoscaling for server utilization and cache performance. In each of these projects, we are particularly focused on an experimental and data-driven approach to the design and verification of such systems.
Our final area of focus is a forward-looking view, examining how the abilities of the network can be extended to offer even greater performance and reliability. Frequently this involves reimagining portions of our architecture, or understanding how we can bring our existing systems together to develop more response systems.
Fundamentally, the Lab’s approach is designed to be pragmatic: focusing on the challenges faced by the systems and infrastructure today, but working towards solving them in ways that open new opportunities in the future. Key to this model is an open collaboration strategy in which we work directly with other engineering teams to best understand the components of the system that they each build and maintain. This approach further keeps the work the team does grounded, and able to produce the most meaningful impacts to the network as a whole.
These collaborations extend beyond just our own organization and include work with numerous researchers in academia. These collaborations have created new opportunities to build relationships with students and academic labs, bringing in fresh outside ideas and approaches. Such collaborations have resulted in participation at operator-oriented venues such as the RIPE meetings and LACNIC, as well as academic venues, such as IMC, PAM, TMA, and others.
Our team is also frequently home to exceptional Ph.D. students for internships. In the past, this has included Stephen McQuistin (University of Glasgow), Lan Wei (University of Southern California), Marc Warrior (Northwestern University), Evita Bakopoulou (UC Irvine), Usama Naseer (Brown University), Shruti Lall (Georgia Institute of Technology), Shaghayegh Mardani (UCLA), and Sayak Nag (UC Riverside). Our interns have explored projects ranging from developing anycast route measurements and optimizations to cache optimizations, to machine learning approaches.
Looking forward, we hope to continue our work on these and related topics, and further hope to share some of the progress and projects we’re working on here in this blog. In the upcoming posts from this series, we’ll present some of our ongoing work, including our findings so far, where we see that work developing in the future, and some of our long term ideas on the most interesting areas of study.
For researchers interested in learning more about Edgio Labs & Advanced Projects, or interested in exploring collaborative works on any of the topics described above, please reach out to the team at firstname.lastname@example.org.
Building out the Basics with Hoplets, Prathy Raman and Marcel Flores, 2021, Passive and Active Measurement Conference (PAM)
Bidirectional Anycast/Unicast Probing (BAUP): Optimizing CDN Anycast, Lan Wei, Marcel Flores, Harkeerat Bedi, John Heidemann, 2020, Traffic Measurement and Analysis (TMA)
Edge Architecture for Dynamic Data Stream Analysis and Manipulation, Orpaz Goldstein, Anant Shah, Derek Shiell, Mehrdad Arshad Rad, William Pressly, Majid Sarrafzadeh (UCLA), 2020, International Conference on Edge Computing (EDGE)
Named Data Networking for Content Delivery Network Workflows, Rama Krishna Thelagathoti, Spyridon Mastorakis, Anant Shah, Harkeerat Bedi, Susmit Shannigrahi, 2020, IEEE CloutNet 2020
Persistent Last-mile Congestion: Not so Uncommon, Romain Fontugne, Anant Shah, and Kenjiro Cho, 2020, Internet Measurement Conference (IMC)
Caching the Internet: A View from a Global Multi-Tenant CDN, Marcel Flores, Harkeerat Bedi, 2019, Passive and Active Measurement Conference (PAM), Puerto Varas
Taming Anycast in the Wild Internet, Stephen McQuistin, Sree Priyanka Uppu, Marcel Flores, 2019, ACM Internet Measurement Conference (IMC), Amsterdam
Distributed Load Balancing in Key-Value Networked Caches, Sikder Huq, Zubair Shafiq, Sukumar Ghosh, Amir R. Khakpour, and Harkeerat Bedi, 2017, Proc. of the 37th IEEE International Conference on Distributed Computing Systems (ICDCS’17), Atlanta, GA
Suffering from Buffering? Detecting QoE Impairments in Live Video Streams, Adnan Ahmed, Zubair Shafiq, Harkeerat Bedi, Amir Khakpour, 2017, IEEE International Conference on Network Protocols (ICNP), Toronto
Peering vs. Transit: Performance Comparison of Peering and Transit Interconnections, Adnan Ahmed, Zubair Shafiq, Harkeerat Bedi, Amir Khakpour, 2017, IEEE International Conference on Network Protocols (ICNP), Toronto
Characterizing Caching Workload of a Large Commercial Content Delivery Network, M. Zubair Shafiq, Amir R. Khakpour and Alex X. Liu, 2016, Proc. of the 35th Annual IEEE Conference on Computer Communications (INFOCOM’16), San Francisco, CA
Optimizing Internet Transit Routing for Content Delivery Networks, Faraz Ahmed, M. Zubair Shafiq, Amir R. Khakpour, and Alex X. Liu, 2016, Proc. of the 24th IEEE International Conference on Network Protocols (ICNP’16), Singapore, Singapore
Riptide: Jump-Starting Back-Office Connections in Cloud Systems, Marcel Flores, Amir R. Khakpour, and Harkeerat Bedi, 2016, Proc. of the 36th IEEE International Conference on Distributed Computing Systems (ICDCS’16), Nara, Japan
QoE Analysis of a Large-Scale Live Video Streaming Event, Adnan Ahmed, Zubair Shafiq, Amir Khakpour, 2016, ACM SIGMETRICS (Extended Abstract), France