People

The handful who built the theory and pushed it into code

The public record of congestion control, queue management, and deliberate bandwidth limitation is surprisingly personal. A small number of engineers, working at a handful of institutions, published the papers that became the RFCs that became the code. Their names appear on the same problems across decades.

The story begins with Van Jacobson. In the late 1980s, Jacobson — then at Lawrence Berkeley National Laboratory in Berkeley, California — looked at a collapsing internet and diagnosed it. The 1988 paper he wrote with Michael Karels, "Congestion Avoidance and Control," delivered at ACM SIGCOMM, described the slow-start and congestion-avoidance algorithms that still run inside every TCP stack. Before that paper, the network had no principled way to back off; senders flooded, routers dropped, and the whole system oscillated. Jacobson's algorithms gave senders a way to probe capacity without destroying it.

Sally Floyd, also working out of Lawrence Berkeley, extended that foundation. Her work on Random Early Detection — RED, developed with Van Jacobson and published in the early 1990s — moved the intelligence to the queue itself. Instead of waiting for a buffer to overflow and then dropping every tail packet, RED would probabilistically drop packets earlier, signaling congestion before it became catastrophic. Floyd was a prolific contributor to the IETF and to ACM SIGCOMM; her published record on fairness, explicit congestion notification, and queue management spans the 1990s and into the 2000s and remains primary source material for anyone studying how queues should behave.

The next large inflection came from Jim Gettys, an engineer with a long history in networking and display systems, who identified and named the phenomenon now called bufferbloat. Gettys noticed that devices — home routers especially, but also handsets and laptops — had accumulated enormous buffers as memory grew cheap. Those buffers filled silently, producing latency spikes of hundreds of milliseconds while reporting healthy throughput numbers. He wrote about it publicly starting around 2010, attracting both collaborators and the attention of the IETF. The name he coined stuck, and the problem he described became a research program.

Kathleen Nichols joined that program and became its most consequential algorithmic contributor. Working with Van Jacobson, she designed CoDel — Controlled Delay — a queue management algorithm that measures the time a packet spends sitting in a buffer rather than counting how full that buffer is. A packet that waits too long is a signal something is wrong, and CoDel acts on that signal. The CoDel specification was published as an IETF informational RFC, and implementations appeared in the Linux kernel's traffic control subsystem. Nichols and Gettys coordinated much of this through Bufferbloat.net, the community hub and clearinghouse that kept the work organized and visible.

Dave Taht came into the same circle as a practitioner and became one of its most energetic builders. His particular focus was getting queue management algorithms onto the constrained hardware that actually handles most consumer traffic — home routers running open firmware. He drove the fq_codel work (fair queuing combined with CoDel) and the Cake algorithm, both of which landed in the Linux kernel and in OpenWrt. Where others published, Taht often shipped.

Luigi Rizzo, a researcher at the University of Pisa in Italy, contributes a different thread. Rizzo designed Dummynet, a traffic shaper that sits inside the kernel's packet filter and can impose delay, loss, bandwidth limits, and queue discipline on any traffic passing through it. Originally written for FreeBSD in the mid-1990s, Dummynet became the machinery that many testing environments — including the one Apple adopted in Network Link Conditioner — rely on underneath. Rizzo published the design in a paper; the tool is still in active use.

What unites these figures is that they published as they went. The 1988 Jacobson and Karels paper, the Floyd and Jacobson RED paper, the Nichols and Jacobson CoDel IETF specification, the Rizzo Dummynet paper — each one explains the mechanism, not just the result. That transparency is why the field accumulated knowledge instead of just accumulated code. Anyone wanting to understand what a shaper is actually doing, or why a queue behaves the way it does, can trace almost every mechanism back to a named person and a legible document.