When I use the tition inference service with opentelemetry the inference performance is much worse than without adding opentelemetry. How to provide inference performance using opentelemtry？

[Leelongqi]

/opt/tritonserver/bin/tritonserver \
  --trace-config mode=opentelemetry \
  --trace-config opentelemetry,url=host:port \
  --trace-config opentelemetry,resource=service.namespace=user_cc \
  --trace-config opentelemetry,resource=service.name=service_a \
  --trace-config opentelemetry,resource=service.version=v8 \
   --trace-config opentelemetry,resource=service.instance.id=998cc493-f310-47de-96bd-71410b7dec09 \
  --trace-config level=TIMESTAMPS \
  --trace-config rate=1 \
  --model-repository=/models \
  --log-error=1

Inferences/Second vs. Client Average Batch Latency
Concurrency: 1, throughput: 58.4353 infer/sec, latency 17099 usec
Concurrency: 2, throughput: 56.4934 infer/sec, latency 35397 usec
Concurrency: 3, throughput: 57.8276 infer/sec, latency 51811 usec
Concurrency: 4, throughput: 57.6041 infer/sec, latency 69392 usec

---

/opt/tritonserver/bin/tritonserver \
  --model-repository=/models

Inferences/Second vs. Client Average Batch Latency
Concurrency: 1, throughput: 133.2 infer/sec, latency 7507 usec
Concurrency: 2, throughput: 156.251 infer/sec, latency 12794 usec
Concurrency: 3, throughput: 156.033 infer/sec, latency 19224 usec
Concurrency: 4, throughput: 156.808 infer/sec, latency 25497 usec

[oandreeva-nv]

I believe the issue is due to the SimpleSpanProcessor which is used at the moment.
There is a PR to fix this: #6733