The sled embedded database is generally regarded as an “ok” choice for working with embedded data in an ergonomic, too-fast-for-its-own-good, transactional manner. But it wasn’t always that way! This talk covers the motivations, design choices, mistakes, and evolution during the first 6 years of this young database’s life. Topics covered: lock-free index structures, low-overhead logging, cheap OLTP transaction techniques, the RUM conjecture’s implications for database design, finding vast troves of bugs with very little testing code in concurrent and stateful systems, considerations around data structure robustness, considerations around composability in stateful infrastructure, and a light dusting of queuing theory’s implications for systems that strive for competitive latency or throughput performance.

This talk is part of the Vaccination Database (Booster) Tech Talk Seminar Series.

Bio:
Tyler Neely is an exiled hermit dwelling in the ruins of what was once a bustling fishing village in north-eastern Germany. While working with large-scale stateful distributed systems in his 20s, he developed a debilitating addiction to debugging in high-stakes situations. The unquenchable thirst for riskier and riskier bugs led him from distributed systems to the lock-free storage engines embedded within them. He spends most of his time anxiously waiting for his fuzzer-driven fault injection systems to spit out his next fix. That leaves a lot of time for hanging out with other lost souls in Berlin, and occasionally building high-reliability low-latency foundational storage infrastructure underpinning the shift to pervasively-distributed edge compute with his company, komora.io.

More Info: https://db.cs.cmu.edu/seminar2022-booster#db5