middlware to track server load #18613
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
an issue was found where load_aware_call could not handle timed out requests. Migrating to use middleware instead while keeping it behind the enable-server-load-tracking flag to avoid any potential performance impact.
with the middleware:
Serving Benchmark Result
Successful requests: 10000
Benchmark duration (s): 166.86
Total input tokens: 1000000
Total generated tokens: 1253399
Request throughput (req/s): 59.93
Output token throughput (tok/s): 7511.73
Total Token throughput (tok/s): 13504.82
---------------Time to First Token----------------
Mean TTFT (ms): 37.11
Median TTFT (ms): 26.30
P99 TTFT (ms): 219.72
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 13.06
Median TPOT (ms): 13.05
P99 TPOT (ms): 13.63
---------------Inter-token Latency----------------
Mean ITL (ms): 13.06
Median ITL (ms): 12.72
P99 ITL (ms): 18.98
without the middleware:
Serving Benchmark Result
Successful requests: 10000
Benchmark duration (s): 170.14
Total input tokens: 1000000
Total generated tokens: 1252755
Request throughput (req/s): 58.77
Output token throughput (tok/s): 7363.02
Total Token throughput (tok/s): 13240.48
---------------Time to First Token----------------
Mean TTFT (ms): 36.96
Median TTFT (ms): 27.08
P99 TTFT (ms): 291.95
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 13.33
Median TPOT (ms): 13.35
P99 TPOT (ms): 13.76
---------------Inter-token Latency----------------
Mean ITL (ms): 13.34
Median ITL (ms): 12.93
P99 ITL (ms): 20.07