Toy systems & networking models
This page is a small, public scratchpad for thinking about systems and networking questions, inspired by work from John Ousterhout and Keith Winstein. Everything here is deliberately simple and illustrative.
toy model
Congestion playground (two flows, one bottleneck)
Very roughly inspired by Remy-style computer-generated congestion control and tools like Mahimahi, this widget shows two long-lived flows sharing a single bottleneck link. The numbers here are made up for intuition; they are not experimental results.
Think “effective throughput” after radio / routing overheads.
Higher values weight Flow A’s delay more heavily in the objective.
“Greedy” heavily favors throughput for Flow A. “Remy-like” nudges the split toward a balance between throughput and delay as you increase the latency sensitivity slider.
Bar height ≈ share of link throughput; darker bars imply more queuing delay. The toy “Remy-like” mode biases the allocation based on the latency weight.
references & reading notes
Systems & networking work this is riffing on
These are some of the papers and tools that motivated this page; the demo above is just a visualization hook to think about their ideas, not a reproduction of results.
- Winstein et al., “TCP ex Machina: Computer-Generated Congestion Control” — congestion-control algorithms synthesized by a program given a network model.
- Netravali et al., “Mahimahi: a lightweight toolkit for reproducible web measurement” — record-and-replay HTTP traffic under emulated network conditions.
- Ousterhout, “The Case for RAMCloud” — large-scale, low-latency storage systems keeping data entirely in DRAM.
- Ousterhout et al., “The RAMCloud Storage System” — design details for microsecond-scale access to petabyte-scale datasets.
- Keith Winstein · Stanford networking and systems group — projects around communication, compression, and computation.