The Fermilab Computing Farms in 2001-2002

al., Merina Albert et

Fermi National Accelerator Laboratory

关键词: Management;    Storage;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;    Monitoring;    43 Particle Accelerators;   

DOI  :  10.2172/811798
RP-ID  :  FERMILAB-TM-2209
RP-ID  :  AC02-76CH03000
RP-ID  :  811798

美国|英语

来源: UNT Digital Library

【 摘 要 】

The Fermilab computing farms grew substantially in 2001 and 2002. This reflected primarily the CDF and D0 computing demand increase as run 2 began and the two detectors and the accelerator performed steadily better, leading to more data and a greater demand for reconstruction computing. In addition, the ''fixed-target'' farms evolved away from the old model of direct tape input and output to a system that uses Enstore and dcache (network-based) as the I/O mechanism. This was part of a more general trend away from a large server used for all of the I/O and many common services and many workers to a model with many smaller systems serving as I/O systems with distributed disk storage. Other major technological achievements include the use of dfarm, a disk caching mechanism, throughout the farms, the upgrades of FBSNG, NGOP for monitoring the farms systems, and generally a more sophisticated management of the machines that constitute these farms. The growth in systems was quite substantial, from a total of 314 dual PCs in early 2001 to 649 duals in early 2003. This does not include the farms that were purchased for CDF and D0 analysis--the CAF and the CAB, nor does it include CMS PC systems. Those systems are sufficiently different in use (and are used differently by the collaborations) that they are treated separately. In future versions of this memo it is likely that all of the systems will be covered.