GMA Instrumentation of the Athena Framework using NetLogger
Craig E. Tull, Dan Gunter, Wim Lavrijsen, David Quarrie, Brian Tierney
Abstract
Grid applications are, by their nature, wide-area distributed applications. This WAN aspect of Grid applications makes the use of conventional monitoring and instrumentation tools (such as top, gprof, LSF Monitor, etc) impractical for verification that the application is running correctly and efficiently. To be effective, monitoring data must be "end-to-end", meaning that all components between the Grid application endpoints must be monitored. Instrumented applications can generate a large amount of monitoring data, so typically the instrumentation is off by default. For jobs running on a Grid, there needs to be a general mechanism to remotely activate the instrumentation in running jobs. The NetLogger Toolkit Activation Service provides this mechanism. To demonstrate this, we have instrumented the ATLAS Athena Framework with NetLogger to generate monitoring events. We then use a GMA-based activation service to control NetLogger's trigger mechanism. The NetLogger trigger mechanism allows one to easily start, stop, or change the logging level of a running program by modifying a trigger file. We present here details of the design of the NetLogger implementation of the GMA-based activation service and the instrumentation service for Athena. We also describe how this activation service allows us to non-intrusively collect and visualize the ATLAS Athena Framework monitoring data.
Create a lesson
Related papers
GC-MoE: Overlapping Expert-Parallel Token Dispatch with FP8/FP4 Grouped GEMM via CUDA Green Contexts
Yong Li, Kan Liu, Tao Lan
When the Consumer Is an Agent: Execution-Grounded Service-Level Agreements
Michal Orzechowski, Renata G. Slota, Jacek Kitowski
PhotonZip: Lifecycle-Aware Compression for HDF5-Based Synchrotron Workflows
Haoxu Li, Wenjing Huang, Chengyu Liu et al.
EpiLoRA: Fusing Many LoRA Adapters into Low-Precision GEMMs for Serving and RL
Hieu Tran, Hong Yu
INSIGHT-HPC: An Interpretable System Intelligence for Global HPC Telemetry
Rupak Kadel, Jie Li, Yong Chen
DynSL: Resource-Adaptive Model Training across Heterogeneous Edge Devices
Boris Radovič, Istemi Ekin Akkus, Marco Canini et al.