Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

HK: Fachverband Hadronen und Kerne

HK 53: Accelerators and Instrumentation II

HK 53.2: Vortrag

Mittwoch, 18. März 2009, 14:15–14:30, H-ZO 90

Dynamic system management for the Alice HLT using the SysMES framework — •Timo Breitner and Camilo Lara — Kirchhoff Institut für Physik, Universität Heidelberg

The task of the ALICE High Level Trigger (HLT) is to reduce incoming detector data from 25 GB/s to at most 1.25 GB/s. This can be achieved by selecting, compressing and filtering the event data in a parallel pipelined chain of analysis processes. The data flow framework used for this purpose supports dynamic runtime reconfiguration of the analysis hierarchy.

System Management in the HLT is performed by the SysMES framework which manages both the computing infrastructure as well as the HLT analysis application. It is able to automatically reconfigure the chain of analysis processes with the following three techniques:

- Monitoring of the infrastructure and application components together with the most critical resources.

- Correlating the monitored values to detect local or global errors

- Execution of actions to return the application into a stable and correct state, in particular to avoid the loss of data. This includes moving, adding and removing processes as needed.

The first version of the HLT application management includes monitors for 18 critical resources and the possibility to react to failures by dynamically reconfiguring the HLT application. Future steps include the development of new more sophisticated rules based on experiences to be gained during the experiment's commissioning phases.