This course aims to teach learners how to design an alternative runtime system called PARTISAN for improved scalability and reduced latency in actor applications. The course covers topics such as dynamic overlay selection, named channels, and affinitized parallelism. The teaching method includes a presentation of the design and implementation of PARTISAN in Erlang, along with demonstrations of its performance improvements. This course is intended for individuals interested in distributed actor systems and network communication optimization.
Overview
Syllabus
Intro
Distributed Actors
Programming Model
Executive Summary We're going to look at how we can improve distributed actor
Limitations: Scalability
Limitations: Latency
Partisan
Dynamic Overlay Selection
Named Channels
Affinitized Parallelism
Experiments
Evaluating Scalability Distributed advertisement counter
Increasing Scalability
Reducing Latency: Microbenchmarks
Increasing Throughput: Echo Service
Increasing Throughput: KVS Service
Takeaways Distributed actor systems limited by implementation assumptions
Questions?
Taught by
USENIX
