Speaker
Dr
Sebastien Binet
(IN2P3/LAL)
Description
`fwk`: a go-based concurrent control framework
============================================
Current HEP control frameworks have been designed and written in the
early 2000's, when multi-core architectures were not yet pervasive.
As a consequence, an inherently sequential event processing design
emerged.
Evolving current frameworks' APIs and data models encouraging global
states, non-reentrancy and non-thread-safety to a more concurrent
friendly environment, in an adiabatic way, is a major undertaking,
even more so when relying on the building blocks provided by `C++`.
This paper reports on the development of `fwk`, a framework
investigating and leveraging the built-in tools of the `Go` language to
enable concurrency at the event- and sub-event-levels. The concurrency
features, code distribution and tooling of `Go` will be first presented
to set the scene and explain why `Go` -and its ecosystem at large- is a
natural fit for a concurrent framework.
The aim for such a framework is to be usable as a big HEP experiment's
control framework. `fwk` is also meant to be a nimble application for
analyses which require quick development/deployment cycles but without
paying the price of a VM on the runtime performances side.
The paper will then discuss the design decisions applied to the
various components of `fwk`, introduce its current capabilities
(I/O, histogramming, dataflow, ...) and highlight how the design
decisions of the `Go` language adequately paved the way for `fwk`.
Then, `fads`, a fast detector simulation toolkit, will be introduced.
'fads' is the port of a single-threaded fast detector simulation
toolkit (`Delphes`) to `fwk`.
The paper will present benchmarks (CPU, RSS, I/O, scalability) of real
`Delphes` use cases against `fads`, as well as comparisons of `fads`
performances with regard to other next-generation concurrent
frameworks, using synthetic data.
Performance tools and concurrency debugging tools provided with the
`Go` toolchain and used for this comparison will also be presented.
Finally, the paper will present prospects and prototypes to fill the
gap in the nascent `Go`-based HEP ecosystem and landscape - namely:
histograms interactive displays and interactive analysis tools.
Primary author
Dr
Sebastien Binet
(IN2P3/LAL)
