Implement Neo4j vs Doublets benchmark in Rust

.github/workflows/rust.yml

@@ -0,0 +1,116 @@
+name: Benchmark Rust version
+
+on: [push, pull_request]
+
+env:
+  toolchain: nightly-2024-01-01
+  GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+jobs:
+  benchmark:
+    runs-on: ubuntu-latest
+
+    services:
+      neo4j:
+        image: neo4j:5.15.0
+        env:
+          NEO4J_AUTH: neo4j/password
+          NEO4J_PLUGINS: '[]'
+        options: >-
+          --health-cmd "wget -q --spider http://localhost:7474 || exit 1"
+          --health-interval 10s
+          --health-timeout 10s
+          --health-retries 10
+          --health-start-period 30s
+        ports:
+          - 7474:7474
+          - 7687:7687
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Setup Rust
+        uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: ${{env.toolchain}}
+          components: rustfmt, clippy
+
+      - name: Wait for Neo4j to be ready
+        run: |
+          echo "Waiting for Neo4j to be fully ready..."
+          for i in {1..30}; do
+            if curl -s http://localhost:7474 > /dev/null; then
+              echo "Neo4j is ready!"
+              break
+            fi
+            echo "Attempt $i: Neo4j not ready yet..."
+            sleep 2
+          done
+
+      - name: Build benchmark
+        run: cargo build --release --all-features --manifest-path rust/Cargo.toml
+
+      - name: Run benchmark
+        working-directory: rust
+        env:
+          NEO4J_URI: bolt://localhost:7687
+          NEO4J_USER: neo4j
+          NEO4J_PASSWORD: password
+        run: cargo bench --bench bench -- --output-format bencher | tee out.txt
+
+      - name: Prepare benchmark results
+        run: |
+          git config --global user.email "[email protected]"
+          git config --global user.name "LinksPlatformBencher"
+          cd rust
+          pip install numpy matplotlib
+          python3 out.py
+          cd ..
+
+          # Create Docs directory if it doesn't exist
+          mkdir -p Docs
+
+          # Copy generated images
+          cp -f rust/bench_rust.png Docs/
+          cp -f rust/bench_rust_log_scale.png Docs/
+
+          # Update README with latest results
+          if [ -f rust/results.md ]; then
+            # Replace the results section in README.md
+            python3 -c "
+          import re
+
+          with open('rust/results.md', 'r') as f:
+              results = f.read()
+
+          with open('README.md', 'r') as f:
+              readme = f.read()
+
+          # Pattern to find and replace the results table
+          pattern = r'(\| Operation.*?\n\|[-|]+\n(?:\|.*?\n)*)'
+          if re.search(pattern, readme):
+              readme = re.sub(pattern, results.strip() + '\n', readme)
+
+          with open('README.md', 'w') as f:
+              f.write(readme)
+          "
+          fi
+
+          # Commit changes if any
+          git add Docs README.md
+          if git diff --staged --quiet; then
+            echo "No changes to commit"
+          else
+            git commit -m "Update benchmark results"
+            git push origin HEAD
+          fi
+
+      - name: Save benchmark results
+        uses: actions/upload-artifact@v4
+        with:
+          name: Benchmark results
+          path: |
+            rust/bench_rust.png
+            rust/bench_rust_log_scale.png
+            rust/out.txt

README.md

@@ -1 +1,48 @@
-# Comparisons.Neo4jVSDoublets
+# Comparisons.Neo4jVSDoublets
+
+The comparison between Neo4j and LinksPlatform's Doublets (links) on basic database operations with links (create, read, delete, update).
+All benchmarks ran with 3000 links in background to increase size of indexes and 1000 are actively created/updated/deleted.
+
+In this particular benchmark we decided not to increase the number of links as Neo4j will not be able to handle it at all in timeframe what GitHub Actions limit allows to use for free. Remember that to get accurate result we ran this benchmark multiple times.
+
+## Task
+
+Both databases used to store and retrieve doublet-links representation. To support storage, and all basic CRUD operations that provide Turing completeness for links as in [the links theory](https://habr.com/ru/articles/895896).
+
+## Operations
+- **Create** – insert point link (link with id = source = target)
+- **Update** – basic link update operation
+- **Delete** – basic link delete operation
+- **Each All** – take all links matching `[*, *, *]` constraint
+- **Each Incoming** – take all links matching `[*, *, target]` constraint
+- **Each Outgoing** – take all links matching `[*, source, *]` constraint
+- **Each Concrete** – take all links matching `[*, source, target]` constraint
+- **Each Identity** – take all links matching `[id, *, *]` constraint
+
+## Results
+The results below represent the amount of time (ns) the operation takes per iteration.
+- First picture shows time in a pixel scale (for doublets just minimum value is shown, otherwise it will be not present on the graph).
+- Second picture shows time in a logarithmic scale (to see difference clearly, because it is around 2-3 orders of magnitude).
+
+### Rust
+![Image of Rust benchmark (pixel scale)](https://github.com/linksplatform/Comparisons.Neo4jVSDoublets/blob/main/Docs/bench_rust.png?raw=true)
+![Image of Rust benchmark (log scale)](https://github.com/linksplatform/Comparisons.Neo4jVSDoublets/blob/main/Docs/bench_rust_log_scale.png?raw=true)
+
+### Raw benchmark results (all numbers are in nanoseconds)
+
+| Operation     | Doublets United Volatile | Doublets United NonVolatile | Doublets Split Volatile | Doublets Split NonVolatile | Neo4j NonTransaction | Neo4j Transaction |
+|---------------|--------------------------|-----------------------------|-------------------------|----------------------------|----------------------|-------------------|
+| Create        | 99201 (0.3x faster)      | 101660 (0.3x faster)        | 85197 (0.4x faster)     | 83516 (0.4x faster)        | 3206352817           | 31808             |
+| Update        | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+| Delete        | N/A                      | N/A                         | N/A                     | N/A                        | 1955576426           | N/A               |
+| Each All      | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+| Each Identity | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+| Each Concrete | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+| Each Outgoing | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+| Each Incoming | N/A                      | N/A                         | N/A                     | N/A                        | N/A                  | N/A               |
+
+## Conclusion
+
+The benchmark results will be automatically updated once the GitHub Actions workflow runs. Results are expected to show that Doublets significantly outperforms Neo4j in both write and read operations due to its specialized data structure for storing doublet-links.
+
+To get fresh numbers, please fork the repository and rerun benchmark in GitHub Actions.

rust-toolchain.toml

@@ -0,0 +1,2 @@
+[toolchain]
+channel = "nightly-2022-08-22"