Nice article. I ran into similar issues when optimizing some code. After looking at a lot of cpu profs I found that llms can be quite powerful, to quickly do micro benchmarks and experiment with potential performance improvements. A lot of dead ends but I found some 10%-20% improvements that way.
Maybe not exactly what you meant but it reminded me about the following: When one of our apple servers failed a decade ago and just vomitted out walls of error logs too fast to read anything,the apple support guy we called took his smartphone and made some photos to read and fix the error.
two things come to my mind:
- use a capacity / slot based model to avoid heavy traffic on small roads
- work together with cities / police etc. to define traffic strategies especially for events.
Shameless, but related, plug for NUNAV (I work for the company behind it, but not on this product): If you are located in Europe you might like to try NUNAV Navigation as well, which uses a collaborative routing based approach to distribute traffic.
Twitter: https://twitter.com/RomPic Email: hn at pickl dot eu Web: https://pickl.eu/