A new processing strategy for the ATLAS event selection system is prop
osed for use downstream of the first level trigger. It aims to achieve
a reduction of the event rate to a level suitable for recording on pe
rmanent storage via a sequential treatment of the ''regions of interes
t'' selected by the level 1 trigger. A possible Trigger/DAQ architectu
re which exploits this concept is presented. It is based on a single p
rocessing farm and a network which transports data and protocol traffi
c. The resources necessary for the implementation of this architecture
are estimated. A set of performance figures obtained by calculations
and computer simulations are given assuming the use of ATM networking
technology.