A Prototype System for the Processor Farm of the HERA-B Experiment at HERA
Paper: 422
Session: B (talk)
Speaker: Gellrich, Andreas, DESY, Zeuthen
Keywords: off-the-shelf products, PC operating systems, data acquisition systems, development environments, trigger systems
A Prototype System for the Processor Farm of the HERA-B Experiment at HERA}
A. Gellrich, R. Dippel, I. C. Legrand, H. Leich, F. Sun, and P. Wegner
DESY - Institut für Hochenergiephysik Zeuthen, Germany
S. Scharein
Humboldt-Universität zu Berlin, Germany
HERA-B Collaboration
For the HERA-B experiment at HERA, starting its physics program in 1998,
a multi-processor system (farm) for the last two levels of a four-level trigger
and data acquisition (DAQ) system is planned.
The main tasks of the processor farm is to run the third level trigger
step and to perform a full event reconstruction already online before finally
selected events are written to mass storage media.
The processor farm will consist of more than 100 nodes and receives event data
via high speed data links from a switching network which performs the event
building.
The event input rate will be up to 500 Hz at an average event size of
70 kbyte.
For costs reasons it is foreseen to use commercially available components as
far as possible.
Based on the experiences gained from building a VME-based test farm system in
1996, a prototype system for the test run period 1997 will be built.
It is supposed to consist of the final building elements of the HERA-B
processor farm which must be ready in 1998.
Main items are the choice of the processor type, offering the
best price/performance ratio, its operating system which must be capable of
running the HERA-B application software, and the connectivity to the HERA-B DAQ
system.
Under discussion for the multi-processor farm are presently, besides standard
DAQ solutions such as embedded VME-based systems, workstation or PC clusters
which run standard operating systems.
First tests have shown that adoption of application software (e.g. the event
reconstruction program) are easy in Unix-like environments, since software is
usually developed under Unix operating systems, and CERN software packages are
used.
In the contribution to the CHEP '97 experiences with the test farm and other
considerations, especially benchmark tests, leading to the concept of the
prototype farm, will be discussed and the set-up of the prototype farm will be
presented.