Speaker
Dr
Sebastien Binet
(IN2P3/LAL)
Description
Current HENP libraries and frameworks were written before multicore
systems became widely deployed and used.
From this environment, a 'single-thread' processing model naturally
emerged but the implicit assumptions it encouraged are greatly
impairing our abilities to scale in a multicore/manycore world.
Thanks to C++11, C++ is finally slowly catching up with regard to
concurrency constructs, at the price of further complicating the
language and its standard library.
Instead, the approach of the Go language is to provide simple
concurrency enabling building blocks, readily integrated into the
language, in the hope that these goroutines and channels will help to
design, write and maintain concurrent applications.
To further investigate whether Go is a suitable C++ and python
replacement language for HENP, we developed go-hist, a multi-threaded
Go package for histogramming.
We will first present the overall design of the go-hist package and
the various building blocks (goroutines, channels and wait-groups)
provided by the Go language which are then leveraged by this library.
A special emphasis will be put on the current SIMD support available
in Go and thus how vectorization can be leveraged in histogramming
code: a cornerstone for automatic performance scalability on the
ever-wider-registers architectures the future has in store.
Then, I/O considerations, such as read/write performances, ease of
serialization and disk sizes will be discussed, for each of the currently
implemented backends (protobuf, gob and json.)
Finally, comparisons with ROOT, Java ROOT and inlib/exlib (an AIDA
implementation in C++) performances will be presented.
Primary author
Dr
Sebastien Binet
(IN2P3/LAL)
