COOKIES: By using this website you agree that we can place Google Analytics Cookies on your device for performance monitoring. |
University of Cambridge > Talks.cam > Computer Laboratory Systems Research Group Seminar > Designing Next Generation Network Interface Cards
Designing Next Generation Network Interface CardsAdd to your list(s) Download to your calendar using vCal
If you have a question about this talk, please contact Srinivasan Keshav. Modern multi-tenant data center applications demand ultra low-latency, very high-throughput communication, and strong performance isolation. Traditional host software stacks cannot keep pace with these demands. In recent years, data center operators have turned to network interface cards (NICs) that implement different types of “offloads” to help accelerate processing, reduce latency, and increase network throughput. However, there are no principled approaches that guide the design of effective offloads, and inform how NICs should be architected to best support tenants’ differing offload needs and ensure isolation. In this talk, I will describe my group’s recent work on building offloads and integrating them into next generation NICs. Using examples, I will present the trade-offs underlying various approaches to dividing the labor between a hardware offload and its corresponding software function. I will then present two multi-tenant offloads that I argue achieve good trade-offs for demanding multi-tenant scenarios: LOOM , which supports hierarchical packet scheduling, and 1RMA, which supports scalable remote memory access. Finally, I will present a new NIC architecture, PANIC , that supports many diverse multi-tenant offloads with programmable traversal and end-to-end isolation. Bio:
This talk is part of the Computer Laboratory Systems Research Group Seminar series. This talk is included in these lists:
Note that ex-directory lists are not shown. |
Other listsHealth and Welfare Research Group IV EURASIAN RESEARCH FORUM 2020 Christ’s College Climate SeminarsOther talksLearning under model misspecification Maximizing Efficiency and Economy of Storage Balance for 100% Renewable Power Supply Variational Bayes as Surrogate Regression Improving EDI in academic and field settings |