Experimental Results of Dynamic Load Scheduling in the CMS Data Acquisition System

Abstract

The online Data Acquisition system of the Compact Muon Solenoid (CMS) experiment at CERN's Large Hadron Collider is designed to collect data corresponding to a single collision of particles, referred to as an event, from about 500 detector Front-Ends. Each of those Front-Ends delivers event-fragments of an average size of 2 KB at a rate of 100 kHz. The event-fragments are statically distributed (usually in round robin fashion) between 8 identical computing farms, which construct the whole events. In this paper we present experimental results of employing a distributed, asynchronous load scheduling algorithm in place of the static event allocation mechanism. The research focuses in particular on balancing the event flow in case of degradations in computing power or network throughput. The discussed studies prove that the proposed method meets the requirements of CMS experiment and has a positive impact on the resource utilization and overall fault tolerance. © Springer-Verlag Berlin Heidelberg 2013.