[Table of Contents] [Previous Section] [Next Section]


Data Acquisition and Computing

The data acquisition architecture for D0 in Run II, as illustrated in Fig. 8, will be largely unchanged from the current system. The basic components will remain: VME Buffer Drivers (VBDs) in each front end electronics crate driving one of eight parallel high-speed Data Cables feeding Multi-Port Memories (MPMs) accessed by a farm of event-building and software-filtering Level 3 processor nodes with another VBD -- Data Cable system feeding event data-logging and monitoring host cluster.

The VBD -- Data Cable -- MPM path will handle a rate of 160 Mbytes per second (~ 800 Hz) to the Level 3 processor farm using existing components. The 48 Level 3 nodes will be replaced with ones running an open operating system, i.e. one which will provide the necessary real-time functions yet will allow great flexibility in choice of commercial hardware. An example Level 3 system in current terms might be a PCI-based PC running Windows NT. The processor farm must possess sufficient compute capacity to provide a software trigger rejection factor of approximately eighty by executing a substantial subset of the current offline algorithms.

The VBD0 -- Data Cable -- MPM path from Level 3 to the host system will be designed to accommodate a rate of 5 Mbytes per second (~ 20 Hz). The host system is also to be assembled from commercial components. Incoming event data will be written to local disk buffers for later spooling to local or remote serial media. The host system will also provide the platform for monitoring of the data stream and will act as the interface to the hardware monitoring and control system.

Events will be reconstructed on the FNAL processor farm system, with that portion dedicated to D0 (an estimated 20,000 MIPs) capable of matching the 10 -- 20 Hz data acquisition rate. Following reconstruction, data will be stored on a tightly coupled disk and robotic tape system, and made available for analysis on a centralized analysis processor. We expect ~250 million events per year to be accumulated, requiring 3 Tb disk-resident and 160 Tb tape-resident storage.


[Table of Contents] [Previous Section] [Next Section]