Some news:
In the most recent edition of SIGCOMM Computer Communications Review, I talk about a radical shift occuring in datacenter server architectures which places network interface cards (NICs) in control of where compute takes place on the server. In this editorial, I both describe the changes that are happening due to SmartNICS and Data Processing Units, as well as open questions for the research community. Read on here.
Many recent computer systems combine new software designs with novel harwdare components. However, comparing two systems which use heterogeneous hardware is fundamentally different from comparing systems running on the same type of hardware. This week at HotNets 2023, Hugo Sadok presents principles for writing research papers comparing the performance of systems using different hardware platforms.
Database proxies intermediate network connections for database management systems (DBMSes) by performing both network-layer and DBMS-specfic offloads. Tigger, presented today at VLDB 2023 by Matt Butrovich, is a new database proxy which uses eBPF to push database functionality into the kernel, improving the efficiency of the proxy. Learn more here about this epic crossover of database and networking research.
Congratulations to Hugo Sadok and the Enso team for winning the Jay Lepreau Best Paper Award and the Distinguished Artifact Award at OSDI 2023! Hugo’s work shows how to improve communication between the network and the CPU to speed up applications I/O intensive applications by 1.5-6X. Read the paper or check out the open-source code.
I was honored to give the opening keynote at the EuroP4 2022 Workshop in Rome. In the talk, I discuss how we should think of SmartNICs not merely as accelerators, but as a radical shift in where the control of compute lies within a server. You can watch a recording of the talk here.
Last Spring, my graduate computer networking students read how the early Internet’s architecture was designed to be resilient to physical damage and attack. With the Russian invasion of Ukraine, CSD MS students Akshath Jain, Deepayan Patra, and Mike Xu saw an opportunity to study how well Internet service had endured under these rare conditions. Today, they presented their findings – a collaboration with Google’s m-Lab – at the 2022 Internet Measurement Conference.
How can a network controller decide which functionality is best deployed on a PISA switch, versus a generic server, versus a smartNIC? The right decision, we argue, is in the eye of the beholder: some operators may optimize for the lowest latency, and others, for the lowest power cost. Today at the Symposium on SDN Research, Francisco Pereira presents on his work-in-progress, SyNAPSE, which aims to distribute network functions across heterogenous hardware. Read more here and watch the talk here. This is a CMU Portugal collaboration.
Algorithmic Complexity Attacks (ACAs) are a powerful class of denial of service attack where a resource-strapped attacker can, using a small amount of attack traffic, nonetheless overload a highly provisioned server. Today at SIGCOMM 2022, Nirav Atre demonstrated a new approach called SurgeProtector that can mathematically guarantee that a system is safe from ACAs. Read a summary of the work from Cylab, dig into the code, or read the research paper to learn more.
When wireless networks suffer sudden throughput degradation, queues build up rapidly at APs and degrade applications like video conferencing. Today at SIGCOMM 2022, Zili Meng shows how wireless APs can quickly signal to senders to slow down transmission when throughput degrades; improving tail network latency and making videos play out smoother. You can read about Zili’s system Zhuge here.
Systems like SurgeProtector and many others rely on heuristics to predict job sizes in order to implement shortest-job-first scheduling. However, heuristics are not always correct which can lead to error in the systems that use them. Today at SIGCOMM 2022, undergraduate Erica Chiang presented a poster on her work on Robust Heuristics for SurgeProtector and took second place in the SIGCOMM Undergraduate Student Research Competition. Congratulations, Erica! Read more on Erica’s work here.
Thank you to Intel for awarding an Intel Outstanding Researcher Award to the Pigasus faculty (James Hoe, Vyas Sekar, and myself) for our work on open-source intrusion detection at 100Gbps and beyond! Read the announcement here.
I am honored and grateful to my alma mater, the University of Washington College of Engineering, for selecting me for their Early Career Diamond Award. Go Huskies!
Internet services today are continuously innovating to provide the best performance for their services, including by developing new Congestion Control Algorithms (CCAs). However, these algorithms are often proprietary or unpublished – leaving the Internet community in the dark whether these CCAs are stable, fair, or efficient. Today at HotNets 2021, Margarida Ferreira presents a prototype called Mister880 that allows researchers to reverse engineer deployed CCAs using program synthesis. Read the paper or watch the talk to learn more!
One year ago at OSDI 2020, we published Pigasus – an FPGA-based IDS and the first open-source intrusion detection system to scale to 100Gbps with a single server. Today Pigasus is open-sourced, used in research at at least 7 universities, and in engagements with 4 companies. To connect this community and share the latest Pigasus-related updates, we started the Pigasus developers’ meeting which met today for the first time! Sign up to join future meetings at mail.pigasus-ids.org.
Congestion Control Algorithms (CCAs) determine important properties like fairness, utilization, and performance stability on the Internet. These properties are well-studied under the assumption that congestion occurs at the `edge’ of the network (such as home network links) – but what happens when congestion occurs in core Internet links as observed in recent studies? Adithya Philip provides some early insights into this question at IMC 2021 today. You can watch his talk or read the paper to learn more!
Benjamin Carleton, who interned with our group through the REUSE program last summer, won the SIGOPS Undergraduate Student Research Competition (SRC) at SOSP 2021! Benjamin’s work was mentored by Nirav Atre and focuses on delayed hits in the context of multi-tier caches. He will go on to represent SIGOPS in the ACM-wide SRC.
In our HotNets 2019 paper, Ranysha Ware discussed why ‘harm’ is a better way of thinking about congestion control deployability rather than ‘fairness.’ The latest Systems Approach textbook now includes a discussion of harm in introducing congestion control to students.
This week, Facebook suffered an all-day outage due to errors in their BGP and DNS configuration. I spoke to Charlotte Hu of Popular Science about what went wrong and why increasing centralization of Internet services makes us more prone to such catastrophic failures. Read more here.
Nirav Atre was awarded CMU’s Cylab Presidential Fellowship in support of his work on denial-of-service attacks and algorithmic complexity. Congratulations, Nirav! Read more here.
I enjoyed talking with Shira Ovide of the New York Times about how the Internet’s constant ability to adapt to failures and congestion kept things running (more or less) smoothly during COVID stay-at-home orders. Read her article here.
Kernel bypass approaches to networking provide very high throughput and low latency, at the cost of ease of administration. Operators lose many debugging, management, and scheduling tools that OS kernels typically offer when packets bypass the kernel altogether. At HotOSXVIII today, Hugo Sadok suggests that we might be able to keep both manageability and high performance by implementing an OS interposition layer on FPGA-based SmartNICs. Read the paper and watch the talk to learn more.
When network operators deploy middleboxes like firewalls and proxies, they often do so to comply with regulations and formal policies. But how can a network operator prove to an auditor that their deployment is compliant with these policies? Grace Liu shows how to do this with a system called AuditBox at NSDI 2021 today.
Every year the National Science Foundation awards fellowships to around 2000 elite graduate students across all areas of STEM, from physics to mathematics to computer science. This year, Minya Rancic – our former intern through CMU’s REUSE program – was awarded one of these prestigious fellowships. Minya will be joining us at CMU for his PhD this Fall. Congratulations!
My spouse and colleague Ruben Martins and I welcomed a new collaborator, Carlos Sherry Martins, to our family on February 27.
FPGAs are rapidly expanding their role in datacenters and a wide range of applications today benefit from them. Today, we launch a new research center spanning CMU, UT Austin, and U. Toronto to explore the future of FPGAs in datacenter server architecture and software. Read more here to learn about the Intel/VMware Crossroads 3D-FPGA Academic Research Center and sign up to follow our seminar series!
9-year-old Maggie and her mom Abby run a podcast called `Big If True’ where they talk about things that are big. The Internet is pretty big, so I joined them this week to talk about just how big it is!
I was overwhelmed and honored to learn that VMware granted me the VMware Systems Research Award for 2020. Thank you so much to VMware and to the kind folks who nominated me!
The Internet Engineering Task Force (IETF) defines the standards that shape how the Internet works. Yesterday, Ranysha Ware presented her work on deployability thresholds for new congestion control algorithms to the Internet Research Task Force at the IETF 109 Meeting.
This week, we saw Zhipeng Zhao’s project Pigasus featured in The Morning Paper and in Dark Reading. In addition, Nirav Atre penned a guest blog post for the APNIC blog about his work on Caching with Delayed Hits.
Intrusion Detection Systems (IDS) are fundamental components of network security infrastructure. Zhipeng Zhao presented our new IDS, Pigasus, today at OSDI 2020. Pigasus can process 100Gbps using a single server, using 38X less energy and two orders of magnitude fewer CPUs than existing open-source approaches. Pigasus is available on Github. Read the paper here or watch the talk to learn more.
Fall Teaching: Information about 15-441/641 is all online at www.myheartisinthenetwork.com. You can schedule office hours to meet with me here.
Textbooks tell us that requests to a cache result in two possible outcomes: hits and misses. Nirav Atre discusses today at SIGCOMM 2020 that there is actually a third possible outcome: a
When network functions share the same server, they achieve slower performance than if they run alone. But why does this happen even when the server is configured to isolate processors and NIC queues? And how can we predict the slowdown? Antonis Manousis explains today at SIGCOMM 2020. You can read the paper here or watch the video to learn more.
Rukshani Athapathu, Tooba Bajwa, and Serena Vincent all presented at the N2Women workshop co-hosted with SIGCOMM 2020. Read about Rukshani’s work on Internet fairness here; Tooba’s work on optimal sketching parameters here; and Serena’s work on NADA congestion control here. Talk videos are available at the workshop homepage.
(Keep an eye out – Rukshani and Serena are applying to PhD programs this Fall!)
I am honored to be named to DARPA’s ISAT Study Group, which aims to identify new areas of development in computer and communication technologies and to recommend future research directions.
I was happy to answer some questions for The Daily Telegraph, Vox Recode, and Pittsburgh’s WPXI about why increased Internet traffic due to COVID19 isolation won’t break the Internet, but might still make videoconferencing harder.
For our HotNets 2019 paper, `Beyond Jain’s Fairness Index’, the IRTF has awarded a 2020 Applied Networking Research Prize to project lead Ranysha Ware. Congratulations, Ray!
Applications to apply to CMU’s REUSE program are now open. This program brings students to CMU to participate in research over the summer, and funds room, board, travel, and a stipend for accepted students. Apply here to intern in my group during Summer 2020.
I am delighted that Roger Yang, our high school intern, has been accepted to CMU to begin his bachelor’s degree! We will be so lucky to have him with us for the next four years.
Ranysha Ware and I were featured in an hourlong interview in the Packet Pushers `Heavy Networking’ podcast. In the interview, Ranysha discusses her research on BBR, deployment of new congestion control algorithms, and how to test whether new algorithms are acceptable to deploy on the Internet. Listen here.
When multiple Internet connections share the same link, they compete for resources. Should connections be required to share the link fairly? If the outcomes are unfair, how unfair can they be? Ranysha Ware presented new guidelines for depoying new congestion control algorithms at HotNets 2019. Read more here.
We’re seeing some media coverage [The Telegraph] [Wired Italian] [Vice] on our BBR findings which were published last week at the Internet Measurement Conference. To clear up some common questions, I’ve written a Medium post in Q&A format.
Ranysha Ware presented our research on modeling the BBR congestion control algorithm at IMC 2019. Watch her presentation on how and why BBR can be unfair to other senders on the Internet here, and read the paper to learn more.
Gizmodo asked me and several Internet experts, ‘What would it take to shut down the entire Internet?’
Current Teaching: Information for 15-441/641 is all online at www.myheartisinthenetwork.com.
My keynote presentation at the Simpósio Brasileiro de Redes de Comunicação e Sistemas Distribuídos is now online. I the talk, I discuss testing and evaluation of fairness for new congestion control algorithms. Watch here in Portuguese.
Google gave our group a Faculty Research Award to study Internet fairness and BBR congestion control. Thank you, Google!
The NSF awarded our group a grant to study fairness on the Internet. Thank you, NSF!
My PhD student Nirav was interviewed about his career path from high school in India to graduate school at CMU.
My PhD student Ranysha was named a Facebook Emerging Scholar. Congratulations, Ray!
Popular Science interviewed me on why websites always seem to break at the worst possible times.
