Speaker
Description
CDAQ: A Distributed and Extensible Data Acquisition Architecture for Nuclear Instrumentation
P. Aguilera$^1$, D. Martínez$^2$, A. Espinoza$^3$, C. Aguilera$^4$, F. Molina$^1$, J. Romero-Barrientos$^1$, M. Zambra$^1$, F. Lopez$^{1,5}$, F. Aguilar$^{1,6}$
$^1$ CEFNEN, Comisión Chilena de Energía Nuclear, Santiago, Chile.
$^2$ Universidad de Concepción, Concepción, Chile.
$^3$ Universidad de Talca, Talca, Chile.
$^4$ Universidad de Santiago de Chile, Santiago, Chile.
$^5$ Universidad Andrés Bello, Santiago, Chile.
$^6$ Universidad Técnica Federico Santa María, Santiago, Chile.
Modern nuclear experiments increasingly rely on computationally demanding acquisition, monitoring, and online-processing pipelines. At the same time, many laboratories and field installations operate with heterogeneous computing resources, including legacy or resource-constrained hardware that cannot easily be replaced or upgraded. CDAQ is being developed as a distributed and extensible data-acquisition framework aimed at exploiting such heterogeneous resources by partitioning acquisition and processing tasks among networked computers. Its architecture builds on concepts established by distributed DAQ frameworks such as CERN's XDAQ and other modular acquisition systems [1–3], while focusing on the use of flexible software components that can be deployed according to the computational capabilities available at each installation.
The objective is not only to distribute the data flow, but to investigate how a DAQ system can sustain long-duration operation using heterogeneous and potentially limited computational resources. Workloads may be distributed between several moderate-performance computers or delegated to more capable local or remote nodes, allowing existing hardware to remain useful while reducing the computational requirements imposed on individual acquisition nodes. For remote or long-duration measurements, CDAQ is also being designed to provide sufficient observability and control to support automatic monitoring, diagnosis, and bounded adaptation during sustained acquisition.
This architecture also provides a testbed for investigating how machine-learning and large language model (LLM) components can be incorporated into modern DAQ systems. Machine-learning algorithms may operate at different stages of the data path for event processing, classification, or anomaly detection [4]. At the supervisory level, we are exploring the use of an LLM-based agent to combine distributed telemetry, logs, configuration, and system state in order to assess the global state of the acquisition system and propose or execute a restricted set of operational decisions. Candidate actions under study include adapting buffer sizes, modifying selected data-flow parameters, or initiating predefined recovery procedures within explicitly constrained limits. Automated DAQ supervision and recovery have already been demonstrated in systems such as CMS DAQExpert [5]; here, the interest is in studying whether LLM-based reasoning can extend this concept toward adaptive supervision of sustained acquisitions while retaining deterministic control over while retaining deterministic control over hardware-critical and time-critical operations..
A working CDAQ prototype currently implements distributed readout, run control, monitoring, and analysis across networked computers and has been integrated with an Ethernet-connected SIS3316 digitizer. Development is now focused on evaluating distributed operation under heterogeneous computational resources and on introducing the mechanisms required for adaptive monitoring and intelligent supervisory services. In this contribution, we will present the architecture and the experimental status reached by the prototype, together with results from the ongoing evaluation of workload distribution and progress toward machine-learning integration and controlled LLM-assisted supervision.
References
[1] J. Gutleber, S. Murray and L. Orsini, “Towards a homogeneous architecture for high-energy physics data acquisition systems,” Comput. Phys. Commun. 153, 155–163 (2003). DOI: 10.1016/S0010-4655(03)00161-9.
[2] E. Lyard et al., “Modern middleware for the data acquisition of the Cherenkov Telescope Array,” Proc. Sci. ICRC2015, 985 (2015). DOI: 10.22323/1.236.0985.
[3] Y. Liu et al., “EUDAQ2—A flexible data acquisition software framework for common test beams,” JINST 14, P10033 (2019). DOI: 10.1088/1748-0221/14/10/P10033.
[4] G. Grosso et al., “Triggerless data acquisition pipeline for Machine Learning based statistical anomaly detection,” EPJ Web Conf. 295, 02033 (2024). DOI: 10.1051/epjconf/202429502033.
[5] G. Badaro et al., “DAQExpert – the service to increase CMS data-taking efficiency,” EPJ Web Conf. 245, 01028 (2020). DOI: 10.1051/epjconf/202024501028.
| Proceedings Publication Intent | Yes |
|---|---|
| Student Award Eligibility | no |
| Financial Support Request | no |