Higher-Performance Stream Processing: Theory and Implementation
We present a portable stream processing library “strymonas” with a familiar declarative interface and very high performance, notably exceeding all competition and attaining the performance of hand-written imperative state machines. The library targets: Java Streams-like applications; digital signal processing including software-defined radio; many sorts of sensor processing such as nested aggregations, compression/decompression.
The library lets us assemble complex stream pipelines just by plugging in simple combinators such as map, filter, flat-map, zip, take, map-accumulate and sliding windowing. Strymonas streams may be infinite, stateful, with the generally variable and statically unknown processing rate.
Strymonas is a code generator with pluggable backends (currently generating C, OCaml and Scala). The pipeline combinators may be freely composed, and yet the resulting convoluted imperative code contains no traces of combinator abstractions: no closures, intermediate objects or tuples. The high-performance is portable and statically guaranteed.
Strymonas has been developed in tandem with the co-algebraic equational theory of stateful streams. The theory lets us convert stream pipelines to normal forms, which are designed to be easily convertible to first-order imperative state-machine code.
http://strymonas.github.io
Venue: OST / Campus Rapperswil / Lecture Hall 3.113
Time: Tuesday, September 23 2025, 17:15-18:00 pm CEST
Speaker: Prof. Dr. Oleg Kiselyov, Tohoku University, Japan
Oleg Kiselyov has been programming and doing research for 45 years, in various languages and in various organizations and companies. Further details about his interests and work can be found on his personal website https://okmij.org/ftp/
Prof. Kiselyov will also be giving a talk on “A Non-Traditional Compilers Course” at the ETH Zürich on Monday 22.09.2025 at 19h15-20h30, and a guest lecture on “Fun with monoids, or map-reduce” at the OST-RJ on Tuesday 23.09.2025 at 10h10-11h40. For further details and registration (optional), please visit https://www.meetup.com/haskellerz/events.
Best regards,
Farhad Mehta