SteNicholas opened a new pull request, #3647: URL: https://github.com/apache/celeborn/pull/3647
### What changes were proposed in this pull request? Introduce `NettyTransportBenchmark` for Netty transport layer performance evaluation. All suites measure performance through the actual Celeborn transport pipeline (TransportServer + TransportClientFactory + TransportContext). Suite overview: 1. RPC Latency - server-client RPC overhead at different payload sizes 2. Concurrent Throughput - multi-client pressure on the transport layer 3. IOMode Comparison - NIO vs native transport (Automatically selects EPOLL/KQUEUE) 4. Server Thread Scaling - validates MAX_DEFAULT_NETTY_THREADS=8 cap 5. Multi-Connection - numConnectionsPerPeer=1 vs 2 vs 4 6. Async Write Pressure - fire-and-forget RPCs to saturate the write path 7. Large Block Transfer - shuffle-like 16MB block transfers (in-memory payload) 8. File-Backed Shuffle - ChunkFetch from disk, NIO vs native transport (EPOLL sendfile bypass detection) Backport: https://github.com/apache/spark/pull/55061. ### Why are the changes needed? Netty is a crucial third-party component for Celeborn. Introduce a micro-benchmark test facilitates: - Verify performance during subsequent Netty upgrades; - Validate performance after changes to relevant code in Celeborn. ### Does this PR resolve a correctness bug? No. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Generated `NettyTransportBenchmark-results.txt`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
