An4kn opened a new pull request, #1584:
URL: https://github.com/apache/ratis/pull/1584

   ## What
   
   Scales the Raft-over-QUIC benchmark from a single machine to a 5-server /
   5-client cluster on SLURM (DCC, Poznan University of Technology) and adds
   the measurements needed to compare the QUIC transport with TCP+TLS.
   
   ## Changes
   
   - RaftBench: closed-loop read-your-writes workload split over several
     client nodes (--worker-offset), control counters (sent / committed /
     failed / connection failures), raw sample dump (--lat-file) merged into
     one percentile pool by scal.sh, and a hop probe that splits latency into
     client->leader and leader->client without clock synchronisation.
   - Leader-side AppendEntries RTT (HOPSTAT) and follower-side gap
     diagnostics (FGAP) printed to server logs.
   - QUIC single-stream layout (raft.quic.server.single-stream) as a control
     variant: same stack, one stream per peer connection instead of one per
     message type.
   - Optional heartbeat thread appender for both transports
     (raft.server.log.appender.heartbeat.thread), plus --rpc-timeout and
     --no-prevote flags for leadership-stability experiments.
   - Cluster scripts: run_matrix.sh (full factorial sweep), matrix6.sh
     (chunking into SLURM reservations), pilot_hb.sh, szukaj_dysk8.sh,
     ssh multiplexing in run_lan.sh, RAM_LOG and READTPUT control 
   - ECDSA P-256 server certificates instead of RSA-4096 (RSA signing cost
     differed between the two BoringSSL builds and skewed QUIC handshakes).
   - Docs: benchmark/README.md, RUNBOOK*, HB-THREAD-CHANGES.md,
     QUIC_IMPLEMENTATION.md updates.
   
   ## Results
   
   72-point matrix (transport x connection model x payload x 5..30 clients),
   summarised in benchmark/README.md section 8.
   
   All new server options default to off; without flags the servers behave
   exactly as before.


-- 
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]

Reply via email to