Speaker
Description
Traditional triggerer DAQ performs explicit hardware rate control: only events passing a fast trigger are fully read out. This reduces throughput and buffering demands, but can lose physics because the decision uses partial detector information and a short time window, and it imposes pipeline buffering on the trigger-latency timescale.
Triggerless (free-running/streaming) DAQ streams all zero-suppressed hits and moves selection to an online reconstruction farm. The key constraint is in-order processing: the farm consumes fixed-width time slices strictly in chronological order, so the event-building network must output one globally time-ordered stream. Variable transport delay creates cross-lane skew; any late slice causes head-of-line blocking and reordering buffer growth, so strict ordering requires an order-preserving barrier that waits for the slowest lane. Thus, the choice and placement of sort/merge primitives directly determine system-wide total buffer size (which can be even unbounded) and sustainable throughput.
We review prior sorter/merger approaches and show why they do not scale to high-rate streaming. We then present a novel architecture: a cascading non-order-preserving merger$\rightarrow$sorter(resequencer) pipeline feeding a single final barrier. Under our scheme, locally bounding skew at each fan-in stage prevents system-wide accumulation, minimizes end-to-end buffering, and maximizes sustained throughput under realistic burstiness and transport variability.
| Minioral | Yes |
|---|---|
| IEEE Member | Yes |
| Are you a student? | Yes |