07-12, 16:30–17:00 (Europe/Amsterdam), REPL (2, main stage)
We discuss the design, development, and debugging of a highly parallel cloud-native application, with a deep dive into the tooling used, while highlighting difficulties and possible improvements within Julia’s ecosystem.
Julia is a great choice for researchers and developers to write expressive and performant code, and promises a smooth transition to production level systems. This talk focuses on that intersection between research and production by explaining a cloud-native application we have built in Julia for highly parallel data processing. We will explain the application's design, demonstrate Julia's strengths, show how we debugged and profiled that application, discuss some of Julia's shortcomings when building such an application, and point out avenues of future work which can address these shortcomings.
Topics include:
- design: AWS, containerization, parallel Julia
- debugging/profiling: Tracy.jl, Intel VTune, Profiling.jl
- work in progress: easier identification of performance impeding lock contention
CTO-Cofounder at Leuko
Software engineering at Leuko.