Date of Award
Doctor of Philosophy (PhD)
Computer and Information Science
The sheer scale of today's data processing needs has led to a new paradigm of software systems centered around requirements for high-throughput, distributed, low-latency computation.Despite their widespread adoption, existing solutions have yet to provide a programming model with safe semantics -- and they disagree on basic design choices, in particular with their approach to parallelism. As a result, naive programmers are easily led to introduce correctness and performance bugs.
This work proposes a reliable programming model for modern distributed stream processing, founded in a type system for partially ordered data streams. On top of the core type system, we propose language abstractions for working with streams -- mechanisms to build stream operators with (1) type-safe compositionality, (2) deterministic distribution, (3) run-time testing, and (4) static performance bounds. Our thesis is that viewing streams as partially ordered conveniently exposes parallelism without compromising safety or determinism. The ideas contained in this work are implemented in a series of open source software projects, including the Flumina, DiffStream, and Data Transducers libraries.
Stanford, Caleb, "Safe Programming Over Distributed Streams" (2022). Publicly Accessible Penn Dissertations. 5194.