Currently, I work as an engineer at Honeycomb.
I also really like climbing rocks and baking pies.
Queueing theory is super boring, right? Wrong! In this talk, we showed how results like the Pollaczek-Khinchine formula and the Universal Scalability Law are actually useful, tractable principles that can help us better design, benchmark, and operate the software we're responsible for.
The internals of operating system and language runtime schedulers are normally opaque to user programs, but can still affect their performance. In this talk at Strange Loop 2016, we unpacked the implementation of the Linux kernel's scheduler, as well as the language runtime schedulers in Go and Erlang. We explored methods for scheduler observability, and discussed some of the performance tradeoffs that scheduler implementations make.
Tools like perf and SystemTap aren't just useful for Linux kernel developers. In this talk at PyBay 2016, we explored some of the problems you can tackle with systems tools, including advanced mixed-mode profiling, memory leak hunting, and call tree generation for better program understanding.