Google Git
Sign in
apache/myfaces/refs/heads/main/./benchmark
tree: 58785ac9f9ca79abab23c77b38eb3e6c9f5bf285
  1. src/
  2. drive.sh
  3. pom.xml
  4. README.md
benchmark/README.md

benchmark — Jetty container benchmark (MyFaces vs Mojarra)

Standalone WAR (NOT part of the myfaces reactor) that runs a few Facelets pages in a real servlet container via the jetty-ee10-maven-plugin, with a PhaseListener recording per-phase wall-clock time. Lets MyFaces 5.0 and Mojarra 5.0.0-M3 be compared on identical pages, and MyFaces before/after an impl change.

Prereqs

Install the MyFaces build you want to measure into the local repo first:

mvn -pl api,impl -am install -DskipTests

(Repeat after each impl change. The -Pmojarra run doesn't need this.)

Run the server

cd benchmark
mvn org.eclipse.jetty.ee10:jetty-ee10-maven-plugin:run              # MyFaces (default)
mvn -Pmojarra org.eclipse.jetty.ee10:jetty-ee10-maven-plugin:run   # Mojarra 5.0.0-M3

Server comes up on http://localhost:8080/ . Pages: table-readonly / repeat-readonly / foreach-readonly / nested-readonly (GET render), and form-inputs (full-lifecycle postback). Stats endpoint: GET /perf-stats (text dump), GET /perf-stats?reset=1 (clear).

Drive it (per-phase timing)

curl URL-range globbing fires many keep-alive requests cheaply:

# warmup
curl -s -o /dev/null "http://localhost:8080/table-readonly.xhtml?w=[1-300]"
curl -s -o /dev/null "http://localhost:8080/repeat-readonly.xhtml?w=[1-300]"
# reset accumulators after warmup
curl -s "http://localhost:8080/perf-stats?reset=1"
# measured run
curl -s -o /dev/null "http://localhost:8080/table-readonly.xhtml?m=[1-3000]"
curl -s -o /dev/null "http://localhost:8080/repeat-readonly.xhtml?m=[1-3000]"
# read per-phase avg_us
curl -s "http://localhost:8080/perf-stats"

Compare the avg_us for RENDER_RESPONSE (and RESTORE_VIEW) per viewId between MyFaces and Mojarra, and between MyFaces before/after an impl change.

Notes / caveats

  • Both impls target Jakarta Faces 5.0 (MyFaces 5.0-SNAPSHOT vs Mojarra 5.0.0-M3), so it's a same-spec comparison. Run both in the same thermal window and take the best-of-N — a busy machine skews results.
  • GET readonly pages exercise RESTORE_VIEW + RENDER_RESPONSE only. form-inputs is a full-lifecycle postback (all six phases); driving it needs a ViewState round-trip and the impl-specific submit marker (form_SUBMIT for MyFaces, form for Mojarra) — submit every hidden field to stay agnostic.
  • PROJECT_STAGE=Production; server-side per-phase time is what's measured, so client/curl speed is irrelevant as long as enough requests are sent.