The active ROB complex

We present results from a modeling exercise allowing to evaluate the impact of grouping readout buffers, with respect to the traffic between individual readout buffers (ROBins) and level-2 processors. The assumption is made that ROBins are grouped into 'active ROB complexes' (AROBCs). An AROBC is assumed to contain a local network of multiple processors, sharing memory and access to a number of ROBins. The modeling does not include the AROBC-internal traffic, and is implementation-independent. Typically, today, an AROBC might be implemented by a commercial SMP system with up to 8 processors and of the order of 20 PCI slots, or by adding substantial computing capacity inside a ROB crate as pursued by the project 'DAQ prototype-1'.