The high energy physics experiment CDF, located in the anti-proton-proton c
ollider at Fermilab, will write data in Run 2 at a rate of 20 MByte/s, twen
ty times the rate of Run 1. The offline production system must be able to h
andle this rate. Components of that system include a large PC farm, I/O sys
tems to read/write data to and from mass storage, and a system to split the
reconstructed data into physics streams which are required for analysis. A
ll of the components must work together seamlessly to ensure the necessary
throughput. A description will be given of the overall hardware and softwar
e design for the system. A small prototype farm has been used for about one
year to study performance, to test software designs and for the first Mock
Data Challenge. Results from the tests and experience from the first Mock
Data Challenge will be discussed. The hardware for the first production far
m is in place and will be used for the second Mock Data Challenge. Finally,
the possible scaling of the system to handle larger rates foreseen later i
n Run 2 will be described. (C) 2001 Elsevier Science B.V. All rights reserv
ed.