Edge-Connected Jaccard Similarity for Graph Link Prediction on FPGA

Abstract

Graph analysis is a critical task in many fields, such as social networking, epidemiology, bioinformatics, and fraud detection. In particular, understanding and inferring relationships between graph elements lies at the core of many graph-based workloads. Real-world graph workloads and their associated data structures create irregular computational patterns that complicate the realization of high-performance kernels. Given these complications, there does not exist a de facto "best" architecture, language, or algorithmic approach that simultaneously balances performance, energy efficiency, portability, and productivity.In this paper, we realize different algorithms of edge-connected Jaccard similarity for graph link prediction and characterize their performance across a broad spectrum of graphs on an Intel Stratix 10 FPGA. By utilizing a high-level synthesis (HLS)driven, high-productivity approach (via the C++-based SYCL language) we rapidly prototype two implementations -a fromscratch edge-centric version and a faithfully-ported commodity GPU implementation -which would have been intractable via a hardware description language. With these implementations, we further consider the benefit and necessity of four HLS-enabled optimizations, both in isolation and in concert -totaling seven distinct synthesized hardware pipelines. Leveraging real-world graphs of up to 516 million edges, we show empirically-measured speedups of up to 9.5× over the initial HLS implementations when all optimizations work in concert.

Electricity Market Dynamics and Regional Interdependence in the Face of Pandemic Restrictions and the Russian–Ukrainian Conflict

“…In most scientific works comparing graph similarity, especially in studies published in recent years, the Jaccard similarity measure was used [34][35][36][37][38][39]. Similarity is essential in scientific research, including machine learning and pattern recognition [40][41][42][43].…”

On the Three P's of Parallel Programming for Heterogeneous Computing: Performance, Productivity, and Portability

“…2) Jaccard similarity: In graph analytics, computing the intersection of neighborhood sets is a widely explored problem [22], [23]. In this work, we evaluate an instance of the set intersection problem to compute link prediction in graph datasets.…”

“…Link prediction in graph datasets can be evaluated using a metric called Jaccard similarity (JS) [24]. Our analysis uses the edge-centric implementation of Jaccard similarity, introduced in [23].…”

“…Algorithm 1 describes our implementation of Jaccard similarity. Similar to prior work on Jaccard similarity for FPGA [23], we use binary search to compute the size of the intersection and union of neighborhood sets. Implementation: Our implementation of Jaccard similarity in OpenCL and SYCL exploits edge-centric parallelism.…”

“…Edge-centric implementation of Jaccard similarity[23] Input : graph G(V, E) in edge-list format:source(|E|), dest(|E|), offsets(|V |) Output: Jaccard(|E|) Data: |E| 1 foreach edge e from E do 2 s = source[e], d = dest[e] // Count neighbors 3 N s = offsets[s+1]-offsets[s] 4 N d = offsets[d+1]-offsets[d] 5 if N s < N d then…”