Control of large scale distributed DAQ/trigger systems in the networked PC
era
This was a one day workshop that occurred inside the NAS2000 workshop. I was
a scientific organizer and presented this paper as well.
Abstract
The HEP
community is moving towards larger and more complex experiments, as exemplified
by the coming LHC ATLAS and CMS detectors. The triggering and data acquisition
hardware has matched this increase in complexity. With availability of cheap
high power PC-based computing, many of the DAQ and higher level trigger
components have been moving to a PC and network based model. These new trigger
and DAQ systems are distributed, sometimes involving 100's of computers. The
control and configuration requirements are larger than anything HEP has seen
previously. These large scale systems must also be fault tolerant: 99% uptime
does not mean all components will be working all the time. However, the
DAQ/trigger cannot stop. The DZERO Run 2 upgrade DAQ and trigger are just such a
distributed and fault tolerant trigger/DAQ system. A sophisticated trigger/DAQ
resource manager has been implemented that can tolerate failures in the system
and even reconfigure the system on the fly. As systems grow, monitoring the
health of the system is more important than ever before. It is no longer
acceptable to present a screen with an indicator light for every single
component in the trigger/DAQ system. Instead, some effort must be made to
intelligently present the information in compact and understandable form for
shift personal. This information must aid them in quick diagnosis of hardware
difficulties (especially those caused by other system the trigger/DAQ depends
upon for operation) as well as early warning of problems. We present the control
system, the monitor system, and the general software design and integration for
the DZERO DAQ and Level 3 trigger, including our design plans for controlling
the expansion of the farm and DAQ bandwidth with minimal work, as well as
the ability for smart monitoring.
Presentations
Links
Research
|
