Fix QUIC echo benchmark: remove --verbose, fix duration overlap and arg serialization#847
Merged
ProjectsByJackHe merged 13 commits intomainfrom May 7, 2026
Merged
Conversation
Point the network-traffic-performance workflow at the merged WinQuicEcho startup ramp fix so quicEcho runs use the new client startup behavior. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Pass the WinQuicEcho repository to the reusable build workflow so the pinned SHA is checked out from the correct repository before building the quicEcho artifact. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Provision msquic-km server certificates in LocalMachine\My instead of CurrentUser\My, clean them up from the matching store, and fail fast when the remote echo_server exits during startup. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Keep remote tool arguments as real token arrays through Invoke-Command and Start-Job instead of collapsing them into a single string token. This lets the remote WinQuicEcho server receive the intended backend, port, and cert-hash arguments. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Treat deserialized PowerShell enumerable token lists such as System.Collections.ArrayList as real argv arrays in the remote invocation helper, instead of collapsing them into a single string token. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Two bugs caused QUIC echo benchmarks over real networks to report dramatically lower throughput than expected: 1. Server exits before client finishes: Both server and client received the same --duration value, but the script waits 5 seconds for server startup before launching the client. With duration=10, the server exits at T+10 while the client runs T+5 to T+15, yielding 0 RPS for the last 5 seconds. Fix: give the server extra duration buffer (startup delay + 5s margin). 2. PS remoting flattens argument arrays: Invoke-ToolInSession passed options via -ArgumentList with (,\) wrapping, but CLIXML serialization still collapsed the array into a single string token. The remote tool received one concatenated argument instead of individual tokens. Fix: serialize options as JSON before passing through -ArgumentList and deserialize on the remote side. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The --verbose flag logs per-datagram events (send state changes, datagram received, datagram echoed) rather than just periodic stats. At high RPS this generates tens of thousands of log lines per second, causing severe I/O backpressure — especially on the server side where output flows through PS remoting CLIXML serialization. Periodic RPS stats print regardless of --verbose. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add PEM cert generation for non-Schannel backends (picoquic, ngtcp2) - Update Run-SendTest/Run-RecvTest to use --cert-file/--key-file for OpenSSL-based backends instead of --cert-hash - Update workflow to build WinQuicEcho with picoquic enabled - Pin WinQuicEcho ref to commit with build_picoquic support Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Alan-Jowett
force-pushed
the
fix/quic-echo-duration-and-args
branch
from
a0c5b76 to
eec1cc6
Compare
ProjectsByJackHe
approved these changes
May 7, 2026
Collaborator
|
LGTM! Let's merge it |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Three bugs caused QUIC echo benchmarks over the real network to report ~9.4K RPS instead of the expected ~3.35M RPS (a 356x undercount).
Bugs Fixed
1. --verbose\ I/O backpressure (224x impact)
The script hardcoded --verbose\ on both client and server. This flag logs per-datagram events (send state changes, datagram received/echoed), not just periodic stats. At high RPS this generates tens of thousands of log lines per second. On the server side, this output flows through PS remoting CLIXML serialization, creating severe I/O backpressure that throttled throughput from 3.35M to ~12K RPS. Periodic \RPS=\ stats print regardless of --verbose.
2. Server duration overlap (halved effective measurement)
Both server and client received the same --duration\ value, but the script waits 5 seconds for server startup before launching the client. With \duration=10, the server exits at T+10 while the client runs T+5 to T+15, yielding 0 RPS for the last 5 seconds. Fix: give the server extra duration buffer (startup delay + 5s margin).
3. PS remoting argument serialization
\Invoke-ToolInSession\ passed options via -ArgumentList\ with (,)\ wrapping, but CLIXML serialization still collapsed the array into a single string token. The remote tool received one concatenated argument instead of individual tokens. Fix: serialize options as JSON before passing through -ArgumentList\ and deserialize on the remote side.
Results
The network result is now 59% faster than the loopback VM benchmark (lab machines have more resources).
Files Changed
Validation