20–23 Oct 2026
University of Washington
America/Los_Angeles timezone

Keeping GPU Tracking-as-a-Service in Sync with Athena Through Shared Memory

20 Oct 2026, 18:50
5m
University of Washington

University of Washington

Alder Hall, 1315 NE Campus Pkwy, Seattle, WA 98105
Poster RDC 5 Trigger & DAQ Reception and Poster Session

Speaker

Owen Huang (University of Washington (US))

Description

Tracking-as-a-service is a way to offload GPU-heavy work to a server when reconstructing tracks in the ITk for the ATLAS Phase-II Event Filter. The design is to have CPU-only Athena clients send hits to a NVIDIA triton server running ACTS traccc algorithm chain. For the reconstructed tracking results to be valid, the client and the server must agree on the detector geometry, conditions and magnetic field settings. At the current stage, we have client-side Athena and the server build their own copies from parsing separate json files. This could lead to silent disagreements between the client and the server. We present the fix to this problem by having Athena as the single source of all the detector inputs. The detector geometry, conditions and magnetic field settings are produced by Athena and allocated into a shared memory region where the server can directly access and read.

Author

Owen Huang (University of Washington (US))

Co-authors

Beojan Stanislaus (Lawrence Berkeley National Lab. (US)) Miles Cochran-Branson (University of Washington (US)) Vakho Tsulaia (Lawrence Berkeley National Lab. (US)) Xiangyang Ju (Lawrence Berkeley National Lab. (US)) Yuan-Tang Chou (University of Washington (US)) Yulei Zhang (University of Washington (US))

Presentation materials

There are no materials yet.