Compare the guarantee before the speed.

One panel measures local power-loss durability. The other measures remote durability on the same RustFS server. Results from these lanes are never pooled.

Campaign status

No validated measurements are published yet.

The participant contract and strict artifact validator are present. A number will appear only after correctness preflight, matched host and power metadata, retained raw samples, and artifact validation.

Panel 1

Local durable commit

This lane compares one mutation per transaction on one writer. Each success includes the participant's qualified local durability barrier. A later reopen verifies every value outside the timed region.

EngineStorage profileEligibilityCurrent result
Flyology.DBFlyology Object Storage Files with power-loss-durable commit policyEligibleAwaiting campaign
SlateDBobject_store LocalFileSystem with fsync enabledEligibleAwaiting campaign
TidesDBUnified WAL and memtable with full sync and no compressionEligibleAwaiting campaign
Panel 2

Remote durable commit on RustFS

Every eligible participant uses the same pinned RustFS image on loopback. A successful sample must include the engine's documented remote-durability barrier, not only submission or immediate visibility.

EngineClient pathEligibilityCurrent result
Flyology.DBIndexed flyology_object_storage=0.1.0-dev clientEligibleAwaiting campaign
SlateDBobject_store AWS client with awaited durable writes and WAL enabledEligibleAwaiting campaign
TidesDBPinned TidesDB S3 pathUnsupportedNot ranked

Why TidesDB is not ranked remotely: the pinned TidesDB implementation can ignore object-store WAL upload failures. A throughput number from that path would not represent the same durable outcome.

Measurement contract

A result must carry its evidence.

The panel accepts a campaign only when the retained artifact names exact engine, adapter, compiler, build flags, dependency source, storage revision, workload, host, power profile, raw samples, errors, and checksum.

  1. Prove eligibilityRun the participant's maintained correctness gate before timing.
  2. Match the environmentMeasure every comparison on one host and one detectable power profile.
  3. Retain raw workKeep every repetition, elapsed time, byte count, error count, and state checksum.
  4. Validate the artifactReject unsupported cells, incomplete provenance, or summaries that do not match samples.
Shared workload

One durable mutation, measured the same way.

Each operation begins one transaction, writes one unique 16-byte key and 1,024-byte value, and waits for the lane's durable commit result. Creation and full close, reopen, read, and SHA-256 verification stay outside the timer.

Writer geometry
One writer, one mutation per transaction, one fresh database per repetition
Timed region
Begin, put, and durable commit
Verification
Close, reopen, read every key, and compare the same state checksum across participants
Campaign geometry
Warmup, operation count, and repetitions are retained inputs, never hidden defaults
Pinned inputs

Source identities stay visible.

The first campaign will resolve Flyology Object Storage from the Flyology Alire index without a Git or filesystem pin. Each result will record both the resolved source commit and the exact Alire-index commit.

Flyology Object Storage
0.1.0-dev resolved fresh from the Flyology Alire index; each campaign records its exact source and index commits
SlateDB
e0161973d8d7ffdede7c44725729838811674e99
TidesDB
23a67a6531bc6c0b537d3696758c7879586dcfce
RustFS
ghcr.io/rustfs/rustfs@sha256:800cf3f352a0a27e3275ca854a51f0027975d7acc7a0d52089a35bcc9fcbf0b5