Data acquisition system issues for large experiments

This talk consists of personal observations on two classes of data acquisition (“DAQ”) systems for Silicon trackers in large experiments with which the author has been concerned over the last three or more years. The first half is a classic “lessons learned” recital based on experience with the high-level debug and configuration of the DAQ system for the GLAST LAT detector. The second half is concerned with a discussion of the promises and pitfalls of using modern (and future) generations of “system-on-a-chip” (“SOC”) or “platform” field-programmable gate arrays (“FPGAs”) in future large DAQ systems. Q system pipeline for the 864k channels of Si tracker in the GLAST LAT consists of five tiers of hardware buffers which ultimately feed into the main memory of the (two-active-node) level-3 trigger processor farm. The data formats and buffer volumes of these tiers are briefly described, as well as the flow control employed between successive tiers. Lessons learned regarding data formats, buffer volumes, and flow control/data discard policy are discussed. ntinued development of platform FPGAs containing large amounts of configurable logic fabric, embedded PowerPC hard processor cores, digital signal processing components, large volumes of on-chip buffer memory, and multi-gigabit serial I/O capability permits DAQ system designers to vastly increase the amount of data preprocessing that can be performed in parallel within the DAQ pipeline for detector systems in large experiments. The capabilities of some currently available FPGA families are reviewed, along with the prospects for next-generation families of announced, but not yet available, platform FPGAs. Some experience with an actual implementation is presented, and reconciliation between advertised and achievable specifications is attempted. The prospects for applying these components to space-borne Si tracker detectors are briefly discussed.