WP 7.5: Backend systems and commercial-off-the-shelf components

Contacts:

  • Alex Keshavarzi, University of Manchester
  • Sophie Baron, CERN
  • Antonio Pellegrino, NIKHEF

Indico link

Project 7.5a: DAQOverflow

The DAQOverflow project aims to provide a benchmark of heterogeneous COTS architectures alongside a open-access, repository-hosted infrastructure and set of commonly used tools and algorithms that will keep pace with evolving COTS technologies (GPU, CPU and FPGA coprocessor farms) for the purpose of cost and performance considered near-detector, near-real-time backend processing for HEP experiments.

Backend processing for HEP experiments has traditionally been the realm of limited localised workflows on FPGAs or dedicated ASICs. In recent years the power and complexity of ASIC devices has increased substantially, and at the same time typically offline workloads have moved closer to the detector using online, near-real-time COTS compute resources (GPU, CPU and FPGA coprocessor farms). The DAQOverflow project aims to keep pace with these COTS technologies as they evolve by benchmarking common TDAQ workflows on a variety of architectures, providing a resource which allows future experiments to pick and choose based on cost- and performance considerations using reference implementations of these workflows.

Contributors: IFIC Valencia (ES), Universidad de Oviedo (ES), University College London (UK), University of Birmingham (UK), University of Bristol (UK), Rutherford Appleton Laboratory (UK), University of Manchester (UK), University of Geneva (CH)

Project contact person: Alex Keshavarzi, University of Manchester

Project 7.5b: From Front-End to Back-End with 100GbE

The perspective of future HEP experiments with lower radiation levels than typically seen at LHC opens the door to increasing the complexity of Front-End electronics, implementing for example RISC-V based processors and SoC in the Front-End. In this context, high throughput 100GbE-based data readout link can reasonably be envisaged. This is a new paradigm which will be investigated in this DRD7.5 Project. It will be tightly linked to other Working Groups like DRD7.2/RISC-V or DRD7.1/links activities.

Streaming data directly from HEP detector Front-End to the DAQ processing farm over Ethernet is very attractive for readout systems of future detectors. Several approaches could be envisaged to reach such a goal: using COTS switches to handle data-streams from the Front-End to Network Interface Cards (NICs) or even DAQ processors (the “No backend” approach), or to design a COTS-based high-density switch bridging the detector environment to the COTS/DAQ world (the “Smart Switch” approach). These approaches are complemented with Back-End board adaptation to explore DAQ topologies with 100GbE (based on the PCIe400 & FELIX for DAQ, concentration and processing) and with the study and design of the building blocks IPs necessary for 100Gb Ethernet cores implementation in future FE ASICs.

These various topics are to be addressed by different collaborators according to their respective expertise.

To address this objective, two main (and complementary) approaches will be investigated in parallel:

  • the “No backend” approach - Theme 1 : using 100GbE COTS switches to handle data-streams from the Front-End to Network Interface Cards (NICs) or even DAQ processors (CERN LBC and ESE groups).

  • the “Smart Switch” approach - Theme 2 : design of a COTS-based high-density switch bridging the detector environment to the COTS/DAQ world (Imperial College).

Their Back-End and Front-End counterparts will be organised as follows:

  • Back-End boards adaptation - Theme 3 : to explore DAQ topologies (based on custom boards for DAQ, concentration and processing) (CPPM CNRS/IN2P3, Nikhef, Brookhaven National Lab (if resource requests are granted in 2024)).

  • Front-End ASICs - Theme 4 : study and design of the building blocks IPs necessary for 100Gb Ethernet cores implementation in future FE ASICs. (Rutherford Lab (if resource requests are granted in 2024)).

Contributors: CERN (CH), CPPM (FR), Nikhef (NL), University of Bristol (UK), Imperial College (UK), Rutherford Appleton Laboratory (UK), BNL (US)

Project contact person: Sophie Baron, CERN and Antonio Pellegrino, Nikhef