Skip to content

README.md

Latest commit

 

History

History
44 lines (33 loc) · 2.73 KB

README.md

File metadata and controls

44 lines (33 loc) · 2.73 KB

Echo: Client/Server Implementations

This folder contains three echo implementations with different networking/concurrency models.

Modules

  • blocking - Blocking socket I/O, benchmarked across platform-thread and virtual-thread server/client combinations.
  • netty - Netty event-loop based non-blocking implementation.
  • selector - Pure Java NIO selector-based implementation.

Final Performance Snapshot (All Echo Submodules)

All benchmark suites use the same load shape:

  • 100 clients
  • 100 messages per client (10,000 total messages)
  • 2 warmups (not measured)
  • 3 measured runs

Numbers below come from each submodule's final benchmark summary examples.

Rank Scenario Average time (s) Throughput (msg/s) Avg (ms/msg) Delta vs fastest
1 blocking: VIRTUAL_SERVER+VIRTUAL_CLIENT 0.116 86128.36 0.0116 +0.000 s (+0.00%)
2 blocking: VIRTUAL_SERVER+PLATFORM_CLIENT 0.132 75544.52 0.0132 +0.016 s (+14.01%)
3 netty: NETTY_SERVER+NETTY_CLIENT 0.206 48483.37 0.0206 +0.090 s (+77.59%)
4 selector: SELECTOR_SERVER+SELECTOR_CLIENT 0.352 28409.09 0.0352 +0.236 s (+203.45%)
5 blocking: PLATFORM_SERVER+VIRTUAL_CLIENT 0.980 10203.37 0.0980 +0.864 s (+744.12%)
6 blocking: PLATFORM_SERVER+PLATFORM_CLIENT 1.790 5586.09 0.1790 +1.674 s (+1441.84%)

Why Metrics Look Like This

  • In the blocking implementation, server thread model dominates total cost; virtual-thread server combinations are much faster because blocked I/O parking is cheaper than running many OS threads.
  • Platform-thread server combinations are slower due to higher OS thread scheduling/context-switch overhead and memory pressure at higher connection counts.
  • Netty performs strongly because one event-loop architecture avoids one-thread-per-connection blocking overhead, but still pays pipeline/event-loop coordination overhead.
  • Selector implementation is also non-blocking and single-loop driven, but manual selector/key/buffer orchestration and single-thread serialized progress can cap throughput earlier than the best virtual-thread blocking path for this specific short-message localhost benchmark.
  • The benchmark is intentionally synthetic and localhost-only; absolute numbers are machine/JVM/OS dependent, but the ordering trend is consistent with the concurrency models.

How To Re-run Final Summaries

mvn -pl echo/blocking/blocking-testing -am test -Dcucumber.filter.tags='@PerformanceSummary'
mvn -pl echo/netty/netty-testing -am test -Dcucumber.filter.tags='@PerformanceSummary'
mvn -pl echo/selector/selector-testing -am test -Dcucumber.filter.tags='@PerformanceSummary'