Add Cloud PDP benchmarks documentation #624
+121
−0
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,120 @@ | ||
| --- | ||
| sidebar_position: 3 | ||
| title: Cloud PDP Benchmarks | ||
| --- | ||
|
|
||
| # Cloud PDP Benchmarks | ||
|
|
||
| This page presents performance benchmarks for the **Cloud PDP**, measured across different policy models, concurrency levels, and relationship depths. | ||
|
|
||
| All tests were run against a production Cloud PDP endpoint over a **10-minute window** per scenario. Latencies are round-trip times (request → response) over HTTPS. | ||
|
|
||
|
Comment on lines
+10
to
+11
|
||
| ## RBAC | ||
|
|
||
| ### Single Concurrency | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 5.80 req/s | 5.60 req/s | 6.00 req/s | | ||
| | Average latency | 8.17 ms | 7.12 ms | 9.80 ms | | ||
| | Min latency | 6.24 ms | 5.90 ms | 6.65 ms | | ||
| | Max latency | 34.12 ms | 7.96 ms | 89.57 ms | | ||
| | P50 | 7.24 ms | 7.11 ms | 7.45 ms | | ||
| | P75 | 7.61 ms | 7.34 ms | 7.93 ms | | ||
| | P90 | 8.50 ms | 7.57 ms | 13.64 ms | | ||
| | P95 | 11.35 ms | 7.81 ms | 23.47 ms | | ||
| | P99 | 20.14 ms | 7.93 ms | 50.17 ms | | ||
|
|
||
| ### 10 Concurrent Requests | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 56.5 req/s | 54.6 req/s | 57.3 req/s | | ||
| | Average latency | 7.80 ms | 6.23 ms | 14.8 ms | | ||
| | Min latency | 4.82 ms | 4.64 ms | 5.1 ms | | ||
| | Max latency | 53.65 ms | 13.13 ms | 97.0 ms | | ||
| | P50 | 6.43 ms | 6.00 ms | 9.9 ms | | ||
| | P75 | 7.82 ms | 6.58 ms | 19.8 ms | | ||
| | P90 | 10.78 ms | 7.22 ms | 33.0 ms | | ||
| | P95 | 14.16 ms | 7.81 ms | 43.6 ms | | ||
| | P99 | 29.94 ms | 10.00 ms | 56.8 ms | | ||
|
|
||
| ## ReBAC | ||
|
|
||
| ### Depth 1 — Single Concurrency | ||
|
|
||
| *Folder → File* | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 5.43 req/s | 5.20 req/s | 5.60 req/s | | ||
| | Average latency | 9.87 ms | 8.90 ms | 12.20 ms | | ||
| | Min latency | 8.04 ms | 7.35 ms | 8.50 ms | | ||
| | Max latency | 28.34 ms | 10.13 ms | 63.80 ms | | ||
| | P50 | 9.05 ms | 8.70 ms | 9.50 ms | | ||
| | P75 | 9.59 ms | 9.12 ms | 12.40 ms | | ||
| | P90 | 10.81 ms | 9.26 ms | 16.90 ms | | ||
| | P95 | 13.32 ms | 9.55 ms | 23.10 ms | | ||
| | P99 | 17.84 ms | 10.00 ms | 45.70 ms | | ||
|
|
||
| ### Depth 1 — 10 Concurrent Requests | ||
|
|
||
| *Folder → File* | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 54.7 req/s | 51.7 req/s | 56.2 req/s | | ||
| | Average latency | 11.3 ms | 9.07 ms | 18.2 ms | | ||
| | Min latency | 6.5 ms | 5.90 ms | 7.8 ms | | ||
| | Max latency | 63.9 ms | 21.22 ms | 258.5 ms | | ||
| | P50 | 9.5 ms | 8.57 ms | 14.5 ms | | ||
| | P75 | 11.7 ms | 9.40 ms | 22.8 ms | | ||
| | P90 | 15.9 ms | 10.81 ms | 33.5 ms | | ||
| | P95 | 20.1 ms | 12.05 ms | 43.0 ms | | ||
| | P99 | 40.0 ms | 13.43 ms | 153.2 ms | | ||
|
|
||
| ### Depth 2 — 10 Concurrent Requests | ||
|
|
||
| *Drive → Folder → File* | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 54.5 req/s | 51.2 req/s | 56.3 req/s | | ||
| | Average latency | 13.3 ms | 10.8 ms | 21.2 ms | | ||
| | Min latency | 7.2 ms | 6.7 ms | 8.1 ms | | ||
| | Max latency | 71.0 ms | 31.7 ms | 103.0 ms | | ||
| | P50 | 10.9 ms | 10.0 ms | 15.7 ms | | ||
| | P75 | 13.7 ms | 11.2 ms | 25.8 ms | | ||
| | P90 | 19.1 ms | 13.6 ms | 40.4 ms | | ||
| | P95 | 26.1 ms | 15.2 ms | 50.2 ms | | ||
| | P99 | 47.5 ms | 18.6 ms | 68.4 ms | | ||
|
|
||
| ### Depth 3 — 10 Concurrent Requests | ||
|
|
||
| *Drive → Folder → File → Comment* | ||
|
|
||
| | Metric | Avg | Min | Max | | ||
| | --- | --- | --- | --- | | ||
| | Throughput | 53.6 req/s | 48.8 req/s | 55.0 req/s | | ||
| | Average latency | 14.5 ms | 11.8 ms | 20.6 ms | | ||
| | Min latency | 8.0 ms | 7.2 ms | 9.2 ms | | ||
| | Max latency | 71.1 ms | 21.6 ms | 125.8 ms | | ||
| | P50 | 12.0 ms | 10.8 ms | 16.4 ms | | ||
| | P75 | 15.1 ms | 12.2 ms | 25.8 ms | | ||
| | P90 | 20.5 ms | 15.0 ms | 36.8 ms | | ||
| | P95 | 26.7 ms | 16.4 ms | 45.0 ms | | ||
| | P99 | 48.6 ms | 18.9 ms | 73.9 ms | | ||
|
|
||
| ## Key Takeaways | ||
|
|
||
| - **Sub-15 ms average latency** across all policy models at the P50 level, even under concurrent load. | ||
|
Contributor
There was a problem hiding this comment. Statistic mislabeled — "average latency" and "P50" are conflated This takeaway reads "Sub-15 ms average latency ... at the P50 level", but average (mean) and P50 (median) are two different statistics. The reader can't tell which one the "Sub-15 ms" claim refers to. Both happen to be true against the tables (P50 Avg ranges 6.43–12.0 ms; Average-latency Avg ranges 7.80–14.5 ms), so pick one and state it cleanly. Suggestion: Reword to a single statistic, e.g. "P50 latency stays under 15 ms across all policy models, even under concurrent load" (or "average latency under 15 ms ..."). (Copilot raised this on this line and it is still unresolved/unaddressed.) |
||
| - **Throughput scales linearly** — 10 concurrent requests yield ~10x throughput with minimal latency increase. | ||
|
Contributor
There was a problem hiding this comment. "Throughput scales linearly" overclaims from two data points The data has only two concurrency levels (1 and 10), so "scales linearly" is stronger than the measurements support — two points can't establish linearity, and the relationship isn't even exactly 10x (RBAC: 5.80 → 56.5 req/s ≈ 9.7x; ReBAC depth-1: 5.43 → 54.7 ≈ 10.1x). Soften to the observed claim. Suggestion: "Throughput scales near-linearly with concurrency — 10 concurrent requests yield roughly 10x throughput with minimal latency increase." (Copilot raised this on this line and it is still unresolved/unaddressed.) |
||
| - **ReBAC depth has modest impact** — adding relationship depth (1 → 3 hops) increases average latency by only ~3 ms under the same concurrency. | ||
|
|
||
| - **Tail latencies stay low** — P99 remains under 50 ms for most scenarios, suitable for latency-sensitive applications. | ||
|
|
||
| :::note | ||
| These benchmarks reflect network round-trip times over HTTPS to the Cloud PDP. | ||
| Actual latency in your application depends on network proximity, payload size, | ||
| and policy complexity. For the lowest possible latency, | ||
| consider deploying the [Edge PDP](/concepts/pdp/overview#run-an-edge-pdp-with-docker) alongside your application. | ||
| ::: | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Benchmark methodology underspecified
For readers to interpret these numbers, the intro should state the measurement conditions: the date/period the run was taken, the Cloud PDP region and the client (load-generator) region/proximity, which endpoint(s) were exercised (e.g.
POST /allow), the request/payload shape, and the policy size (number of roles/resources/relationship tuples). Without these, "5.80 req/s at single concurrency" is hard to reconcile (a single-client closed-loop run is throughput = 1 / latency, which the 8.17 ms avg ≈ 5.8 req/s confirms — worth stating explicitly so it doesn't read as a throughput ceiling).Suggestion: Add a short "Methodology" paragraph under the intro listing date, regions, endpoint, payload, policy size, and load model (closed-loop vs open-loop).
(Copilot raised this on the intro line and it is still unresolved/unaddressed.)