Speaker
Description
Baikal-GVD is a gigaton-volume neutrino observatory under construction in Lake
Baikal. Its data processing software consists of a core part and a managing
part. The former is a set of C++ programs built upon the BARS (Baikal Analysis
and Reconstruction Software) framework, which provides a basis for
implementing all data processing stages. The Python-based management layer
organizes these programs into an executable processing graph, resolving
dependencies between them.
The system is designed with three levels of parallelism. First, different
detector clusters are processed independently on separate virtual machines
(VMs), with a dedicated VM merging their results. Second, each cluster data
processing is performed in two sequential workflows: Fast Processing (per-file
analysis within 2–13 minutes) and Offline Processing (full-run reprocessing
with higher precision, completed within 1–5 hours after a run). Each workflow
contains parallel processing sections.
This work focuses on the third level of parallelism: multi-threading within
the core BARS framework. The use of multi-threading allows for a sizeable
reduction of the algorithms' execution time, which can be particularly
beneficial for online data analysis tasks in the context of multi-messenger
astronomy.
