[jira] [Created] (OAK-12426) Benchmark CacheLIRS and Caffeine for oak-store-document

["Rishabh Daim (Jira)" <ji...@apache.org>]

Rishabh Daim created OAK-12426:
----------------------------------

             Summary: Benchmark CacheLIRS and Caffeine for oak-store-document
                 Key: OAK-12426
                 URL: https://issues.apache.org/jira/browse/OAK-12426
             Project: Jackrabbit Oak
          Issue Type: Task
            Reporter: Rishabh Daim
            Assignee: Rishabh Daim


h3. Summary
Add reproducible performance benchmarks comparing CacheLIRS and the Caffeine-backed Oak cache for oak-store-document before Caffeine becomes the default under OAK-12425.

h3. Scope
* Add the benchmark to {{oak-benchmarks}} so benchmark-only code is reviewed and run independently from the production default switch.
* Exercise representative DocumentNodeStore cache workloads through production cache APIs, including steady-state reads, churn/eviction, invalidation, and concurrent access.
* Compare throughput, latency, hit/miss rate, eviction behavior, and loader/backend-read count for CacheLIRS and Caffeine with identical cache weights and deterministic workloads.
* Include warm-cache and cold-cache scenarios and multiple working-set-to-cache-size ratios.
* Document an exact command, JVM settings, workload parameters, and a concise result table.

h3. Reference
Use the benchmark structure and reporting patterns from [apache/jackrabbit-oak PR 2898|https://github.com/apache/jackrabbit-oak/pull/2898], adapting them to oak-store-document caches rather than segment caches.

h3. Acceptance Criteria
# Both CacheLIRS and Caffeine run against identical deterministic DocumentNodeStore cache workloads.
# Results report throughput/latency plus hit, miss, eviction, and backend-load counts.
# Benchmarks cover steady-state, cache churn, invalidation, and concurrent access with at least three working-set/cache-size ratios.
# The benchmark can be selected from BenchmarkRunner and its run command and JVM requirements are documented in code or existing benchmark documentation.
# Results demonstrate whether Caffeine is at least as performant as CacheLIRS for the representative workloads; any regression is quantified and investigated.
# Benchmark changes are delivered in a PR separate from OAK-12425.




--
This message was sent by Atlassian Jira
(v8.20.10#820010)