GRM-RS reference implementation

Preliminary performance data

Secured gRPC service measurements presented with their environment, methodology, variability, and claim boundaries intact.

Benchmark reports

Secured service baselines

1 / 2

Default report · DigitalOcean London

Median of three secured VPS runs

The median of three independent Criterion run medians on consistent cloud infrastructure.

Publication candidate

3 independent runs

11 typed operations

mTLS local gRPC on VPS

What was measured

Each value is the median of the latency medians reported by three independent Criterion runs. The deterministic binary workspaces contained 250 or 1,000 rows, and every measured operation used an established mutual-TLS connection.

Read, traversal, explain, and profile rows are representative secured VPS measurements. Write rows are indicative because each iteration performs substantial untimed workspace setup.

Aggregate medians

Operation	250 rows	1,000 rows
Edge create indicative	1,330.40 µs 20.9% spread	1,604.54 µs 18.9% spread
Edge endpoint lookup	261.10 µs 6.2% spread	284.57 µs 8.4% spread
Edge update indicative	1,395.00 µs 12.4% spread	1,357.35 µs 12.5% spread
Explain `edge.find`	310.37 µs 16.9% spread	302.85 µs 6.0% spread
Explain `node.find`	349.38 µs 11.9% spread	341.57 µs 5.1% spread
Node create indicative	1,362.10 µs 12.8% spread	1,452.40 µs 22.8% spread
Node property lookup	269.99 µs 11.6% spread	295.41 µs 9.2% spread
Node update indicative	1,232.07 µs 4.5% spread	1,270.65 µs 9.6% spread
Profile `edge.find`	304.34 µs 0.5% spread	306.71 µs 8.4% spread
Profile `node.find`	398.82 µs 11.7% spread	361.96 µs 21.5% spread
Selective traversal	306.77 µs 12.1% spread	304.70 µs 13.0% spread

Spread is (maximum run median - minimum run median) / aggregate median. Setup, startup, connection establishment, certificate generation, and TLS handshake are excluded.

Server profile

Provider / region: DigitalOcean / lon1
Instance: basic-premium-amd-2vcpu-4gb
Machine: 2 logical CPUs, DO-Premium-AMD, 3.8 GiB RAM
Storage: Droplet local NVMe SSD
Platform: Ubuntu 24.04.3 LTS, kernel 6.8.0-71-generic
Toolchain: Criterion 0.5.1; rustc 1.88.0
Source: f8cd87dd49c191d28a37f866c94f60a1852cffaf
Source runs: 10 June 2026 at 07:37, 07:54, and 08:08 UTC

Claim boundary

This supports a reproducible GRM secured-service baseline on the recorded VPS profile. It does not support claims of database superiority, hosted durability, multi-writer behavior, authorization, or production certificate lifecycle.

Original report · Local workstation

Single-run local preview

Criterion median estimates from retained artifacts on a local development workstation.

Exploratory preview

1 retained run

7 typed operations

mTLS local gRPC service

What was measured

Criterion median estimates for seven operations against deterministic 250-row and 1,000-row binary workspaces. Operations used an established mutual-TLS connection.

Fixture preparation, service startup, connection establishment, certificate generation, and the TLS handshake are outside the measured duration.

Criterion medians

Operation	250 rows	1,000 rows
Edge endpoint lookup	126.49 µs 95% CI 115.73-138.79	127.39 µs 95% CI 122.64-142.73
Explain `edge.find`	151.21 µs 95% CI 145.43-166.30	148.74 µs 95% CI 145.67-154.26
Explain `node.find`	184.76 µs 95% CI 175.58-186.60	177.71 µs 95% CI 173.12-185.51
Node property lookup	131.28 µs 95% CI 126.84-134.80	128.63 µs 95% CI 124.16-132.58
Profile `edge.find`	159.32 µs 95% CI 153.38-167.50	154.75 µs 95% CI 148.70-163.86
Profile `node.find`	215.70 µs 95% CI 209.36-231.67	212.16 µs 95% CI 207.77-223.36
Selective traversal	151.64 µs 95% CI 143.58-157.37	152.52 µs 95% CI 148.04-153.35

Run provenance

Completed: 9 June 2026, 15:17 UTC
Dataset: Deterministic 250-row and 1,000-row fixtures
Machine: 12 logical CPUs, Intel Core i7-10750H, 30.2 GiB RAM
Platform: Ubuntu 26.04 LTS, kernel 7.0.0-22-generic, local SSD
Source: c1a9787f5b4ea4e19f11bfa7b5f7b41fa719c3bc (dirty)
Run ID: 20260609T151741Z-local-grpc-mtls-soml-website-preview

Claim boundary

This is an exploratory local preview. It does not support claims of database superiority, hosted durability, multi-writer behavior, authorization, production certificate lifecycle, remote-service latency, throughput, or larger-scale performance.

Preliminary performance reporting

Performance data is useful early when its boundary travels with it. Embedded execution, local services, secured VPS services, and remote environments answer different questions and should not be collapsed into one headline number.

The cloud aggregate is now the primary baseline because it combines three independent runs on a recorded server profile. It remains a latency study, not a throughput test or database comparison. Its two fixture sizes do not establish scaling behavior.

Future reports will retain source revision, dataset shape, persistence format, security mode, server profile, variability, and exclusions. Comparative claims require like-for-like workloads, appropriate indexing, repeatable environments, and multiple runs.