Skip to content

[Feature] add CUDA support in the metrics service #212

Description

@LorenzoTettamanti

Is your feature request related to a problem?

  • Yes

    Problem Description:

    As the documentation and the README announced, the GPU support will be introduced in the 0.1.6 version. GPU's are an essential part of Machine learning and AI so a support to trace CUDA calls must be introduced

    Solution:

  • Introduce a GPU support using uprobes and uretprobes directly in the libcudart.

  • Trace essential functions calls like cudaFree,cudaMalloc,cudaMemcpy

  • Add new OTEL metrics (must be discussed together)

  • Adapt the existing consumer library to consume the GPU extracted data correctly in the pipeline

    Additional info

    common/src/otel-metrics.rs: Contains all the OpenTelemetry metrics instrument to process the data from the eBPF perf buffers
    common/src/semantics.rs: Contains all the semantic conventions for the metrics service
    common/src/consumer.rs: Consumer library to read to consume the data that comes from the perf-buffers


Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

enhancementNew feature or requestrustPull requests that update Rust codev0.1.6

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions