Comparison: OpenTelemetry Collector vs. Prometheus Agent Mode for Lightweight Metric Collection (Benchmark)

[Radhe1002]

Hi Team,

I am trying to determine the more lightweight solution for metric collection and remote writing to a TSDB in terms of memory and CPU usage (benchmark). I have the following two setups:

OpenTelemetry Collector Configuration:

Receiver: Prometheus receiver (scraping metrics from a target)
Exporter: Prometheus Remote Write exporter

pipelines:
metrics/otel-pipeline:
receivers:
- prometheus
exporters:
- prometheusremotewrite

Prometheus Agent Mode Configuration:
Minimal Prometheus configuration for scraping and remote writing:

scrape_configs:

• job_name: 'example-job'
  static_configs:
  • targets: ['example:8080']
    remote_write:
• url: 'http://remote-tsdb:9090/api/v1/write'

Question:
Which of these setups is more lightweight in terms of memory and CPU consumption?
Are there any benchmarks or best practices available to decide between OpenTelemetry Collector and Prometheus Agent mode for this use case?
Does the addition of components like batch/memory limiter in OpenTelemetry pipelines significantly affect resource usage compared to Prometheus Agent?

Environment:

OpenTelemetry version: 0.106.1
Prometheus version: prom/prometheus:2.45.0
Deployment method: Kubernetes

I would appreciate any guidance or insights on this!

[welcome]

Thanks for opening this issue. A team member should give feedback soon. In the meantime, feel free to check out the contributing guidelines.