Skip to content

Benchmarks: Neo4j and Elasticsearch memory settings follow their vendors' guidance #460

Closed

Description

@tae898

Every engine in the benchmark harness runs as its vendor recommends. The served -Xms work (#456, DECISIONS #200) checked the JVM comparators against their vendors' guidance and found two more mismatches beyond the initial heap.

Neo4j: the harness pins the heap at 50% of the cell's memory cap and sets the page cache to the cap minus heap minus 1 GiB. Neo4j's own neo4j-admin server memory-recommendation, run from the pinned image, disagrees at every cap (8g: heap 3600m, page cache 2g; 16g: 5g and 7g; 32g: 12g and 12g; 48g: 18g and 20g). The manual says to use that command for the recommendation and to set the initial and maximum heap to the same value.

Elasticsearch: the docs say the heap should be no more than 50% of the node's memory and below the compressed-oops threshold (26GB is safe on most systems). Every tier the harness runs is at 50% except deep10m, which runs 24g in a 36g cap (67%).

Change: Neo4j arms take the tool's heap and page cache for each tier cap from one table in code, generated from a committed copy of the tool's output and checked by a test; Elasticsearch deep10m moves to 18g with -Xms equal to -Xmx. Docs and method text follow. Existing rows are not edited.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions