VLDB 2026 Artifact - SHARP

Artifact for the VLDB 2026 Submission - Sharp: Shared State Reduction for Efficient Matching of Sequential Patterns

This repository provides the artifact for Sharp, a system for efficient best-effort pattern matching using shared state reduction. It supports three workloads: CEP/ for Complex Event Processing, MATCH_RECOGNIZE/ for SQL-based row pattern matching, and GraphRAG/ for path-based pattern matching over knowledge graphs. Sharp leverages pattern-sharing and a lightweight cost model to significantly reduce computational overhead while preserving high recall under latency constraints.

CEP and MATCH_RECOGNIZE Experiments

The codebase has been tested on Ubuntu 22.04, SUSE Linux Enterprise Server 15 SP5, and Red Hat Enterprise 9.5. For both CEP/ and MATCH_RECOGNIZE/, enter each directory and install the necessary dependencies listed in build_support/packages.sh:

$ sudo build_support/packages.sh
$ sudo apt install libboost-all-dev

Download Data

Download the required datasets from synthetic dataset, real-world datasets.
Create the following directories and unzip the datasets inside accordingly:

$ mkdir synthetic_data real_data
# unzip downloaded files into the above folders

Compile the Code

Inside both CEP/ and MATCH_RECOGNIZE/:

$ mkdir build && cd build
$ cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=YES -DCMAKE_BUILD_TYPE=Debug ..
$ make -j$(nproc)

Run Experiments

Navigate to the scripts/ folder, set up a Python virtual environment, and install dependencies:

$ python -m venv venv && source venv/bin/activate
$ pip install pandas matplotlib
$ python recall_latency_throughput_parallel.py

---

GraphRAG Experiment

Navigate to GraphRAG/GraphRAG-SHARP/, create a virtual environment, and install dependencies:

$ python -m venv venv && source venv/bin/activate
$ pip install -r requirements.txt

Download Data

The dataset can be downloaded here.

Inference Pipeline

Ensure access to a GPU (≥12GB). Set your Hugging Face token:

$ export HF_TOKEN="<TOKEN>"

Step 1: Generate Path Queries

$ ./scripts/planning.sh

Step 2: Generate Answers with LLM

After step 1 completes:

$ ./scripts/reasoning.sh

Repeat this process for each baseline folder.