Sim Compare

• For recency-heavy: linked.S4Lru, adaptive.Arc
• For frequency-heavy: linked.Lfu, irr.Lirs
• For scan-resistant: two-queue.TwoQueue, two-queue.S3Fifo
• For size-aware traces: greedy-dual.Gdsf, greedy-dual.Camp

Choose cache sizes. Use a geometric progression covering the working set:

• Start small (e.g., 100), end near working set size
• 5-8 sizes: e.g., 100,500,1_000,2_500,5_000,10_000,25_000
• If the trace has few distinct keys, reduce the range

Run the simulation. Use the Gradle task:

./gradlew simulator:simulate -q \
  --maximumSize=100,500,1000,2500,5000,10000 \
  --metric="Hit Rate" \
  --title="Description" \
  --theme=light \
  --outputDir=build/reports/sim

Override the trace and policies via system properties appended to the command:

-Dcaffeine.simulator.files.paths.0="format:path/to/trace"
-Dcaffeine.simulator.policies.0=product.Caffeine
-Dcaffeine.simulator.policies.1=opt.Clairvoyant
# ... etc

Note: for single-size runs, use ./gradlew simulator:run -q with -Dcaffeine.simulator.maximum-size=N instead of simulator:simulate.

Read and interpret results. The simulate task produces:

• Individual CSV per cache size
• Combined CSV (policies as rows, sizes as columns)
• PNG chart (line graph of metric vs cache size) Read the CSV output files in the output directory:
• Compare hit rates across policies at each cache size
• Identify the crossover points where one policy overtakes another
• Note the gap between Caffeine and Clairvoyant (theoretical ceiling)

Explain findings. For each notable result:

• WHY does policy X beat policy Y on this trace?
• What trace characteristic drives the difference? (frequency bias, recency bias, scan patterns, temporal shifts)
• How close is Caffeine to optimal? Where does it lose?
• Reference the relevant research paper if applicable:
  • TinyLFU paper for admission filter behavior
  • Adaptive paper for hill climber effectiveness
  • See .claude/docs/research-foundations.md for paper-to-code mapping

Report. Present:

• Summary table of hit rates at each cache size
• Key takeaways (2-3 sentences)
• Notable policy behaviors
• Path to generated chart PNG