AMBER Experiment’s Online Filter System for Virtualized IT Infrastructure

Abstract

The operation of high-level trigger (HLT) systems in high-energy physics experiments requires the utilization of substantial computing resources. Typically, these systems are constructed as computing farms with cutting-edge expensive hardware to provide sufficient computing power. Usually the systems are situated on-site and process detector data in real-time to minimize latency. This article presents an alternative high-level filter system designed for the AMBER experiment at CERN. The novel aspect of our approach is its high efficiency, which removes the necessity for a dedicated on-site computer farm. Instead, it makes use of existing shared resources located within the CERN data center. The proposed system is capable of efficiently handling the data generated by the medium-sized experiment and performing numerous parallel filtering tasks in real time. All system components operate within a shared, fully virtualized environment, including databases, storage, and processing units. This flexible environment scales effectively, allowing for adjustments to allocated resources in accordance with agreements with service managers. We present the architectural design and the implementation of such a system. To demonstrate its capabilities, we have conducted a series of various measurements assessing its performance, latencies, and stability under maximum (expected) loads. The results demonstrate the resilience and reliability of the filtering system while optimizing overall costs to a minimum.