Demystifying performance of eBPF network applications
Guest Post: eBPF has been widely leveraged to improve network function performance. Can similar benefits be achieved for web servers and microservices?
Tag: networking
Centralized or distributed? Understanding modern router and AI fabric architectures
Guest Post: How distributed forwarding and modern data plane architectures enable routers and AI fabrics to scale reliably and handle massive workloads without single points of failure.
21 years and counting of ‘eight fallacies of distributed computing’
Eight long held and common beliefs about the network have been shown, time after time, to be false. What are they, and what do they mean?
Flow control in switch ASICs: Part 3 — Notes on flow control
Guest Post: Flow control is a fundamental concept in the design of high-performance networks. In part three of this three-part deep dive series, we explore how flow control is implemented in switch ASICs.
Rate-based versus credit-based flow control: Part 2 — Notes on flow control
Guest Post: Flow control is a fundamental concept in the design of high-performance networks. In part two of this three-part deep dive series, we compare rate-based and credit-based approaches to flow control.
Internet in Spaaaaace
Marc Blanchet’s presentation about deep space networking imagines a future where the Internet works well for people on the moon, and those they are connected to back on Earth.
Principles of flow control: Part 1 — Notes on flow control
Guest Post: Flow control is a fundamental concept in the design of high-performance networks. In part one of this three-part deep dive series, we look at the principles of flow control.
‘Shine a light’ — ending the exile of being too far from Main Street
Free-air laser communications could overcome path problems confronting network infrastructure and help address the ‘last mile’ delivery problem.
Why your AI factory’s biggest bottleneck isn’t the GPU — it’s the network
Guest Post: AI breakthroughs aren’t just about GPUs — networks are the bottleneck. Even 0.1% packet loss can cut GPU use by 13%, making job completion hinge on robust, predictable network design.
Estimating the sweet spot of packets for classifying network flows at scale
Guest Post: Developing FastFlow — a time-series method that efficiently classifies network flows using the fewest packets needed for accurate results.