The Social Network Benchmark (SNB) suite defines graph workloads targeting database management systems and is maintained by the LDBC SNB Task Force (work charter).
The benchmark suite consists of two distinct benchmarks on a common dataset, since there are two different workloads:
- Business Intelligence workload is focusing on aggregation- and join-heavy complex queries touching a large portion of the graph with microbatches of insert/delete operations. Its data sets are available in Cloudflare R2.
- Interactive workload captures transactional graph processing scenario with complex read queries that access the neighbourhood of a given node in the graph and update operations that continuously insert new data in the graph. Its data sets are available in the CWI/SURF data repository.
Each workload produces scoring metrics for performance at the given scale and price/performance metrics.
The Full Disclosure Report (FDR) further breaks down the composition of the metric into its constituent parts, e.g. single query execution times.
For a brief overview, see our talk given at FOSDEM 2023’s graph developer room. The Social Network Benchmark specification can be found on arXiv. If you are interested in getting audited results, please reach out at info AT ldbcouncil.org.
SNB Interactive workload – Audited results
| Benchmark setup | SF | Hardware | System cost | Throughput | Documents |
|---|---|---|---|---|---|
|
30 | Alibaba Cloud ecs.g8y.16xlarge, 256GiB RAM, 64×Arm-based YiTian 710 |
277,542 RMB | 16,133.08 ops/s | |
| 100 | Alibaba Cloud ecs.g8y.16xlarge, 256GiB RAM, 64×Arm-based YiTian 710 |
277,542 RMB | 16,966.26 ops/s | ||
| 300 | Alibaba Cloud ecs.g8y.16xlarge, 256GiB RAM, 64×Arm-based YiTian 710 |
277,542 RMB | 13,532.62 ops/s | ||
|
30 | AWS r5d.12xlarge, 384GiB RAM, 48×Intel Xeon Platinum 8259CL |
291,176 USD | 12,252.50 ops/s | |
| 100 | AWS r5d.12xlarge, 384GiB RAM, 48×Intel Xeon Platinum 8259CL |
291,176 USD | 12,934.61 ops/s | ||
| 300 | AWS r5d.12xlarge, 384GiB RAM, 48×Intel Xeon Platinum 8259CL |
291,176 USD | 12,721.24 ops/s | ||
|
30 | AWS r5d.12xlarge, 372GB RAM, 48×Intel Xeon Platinum 8259CL |
263,282 USD | 9,285.86 ops/s | |
| 100 | AWS r5d.12xlarge, 372GB RAM, 48×Intel Xeon Platinum 8259CL |
263,282 USD | 8,501.21 ops/s | ||
| 300 | AWS r5d.12xlarge, 372GB RAM, 48×Intel Xeon Platinum 8259CL |
263,282 USD | 8,370.52 ops/s | ||
|
30 | AWS r5d.12xlarge, 374GB RAM, 48×Intel Xeon Platinum 8175M |
280,650 USD | 5,436.47 ops/s | |
| 100 | AWS r5d.12xlarge, 374GB RAM, 48×Intel Xeon Platinum 8175M |
280,650 USD | 5,010.77 ops/s | ||
| 300 | AWS r5d.12xlarge, 374GB RAM, 48×Intel Xeon Platinum 8175M |
280,650 USD | 4,855.52 ops/s | ||
|
30 | AWS r6id.8xlarge, 256GiB RAM, 32×Intel Xeon Platinum 8375C, 1 read thread, 1 write thread |
216,222 USD | 3.04 ops/s | |
| 30 | AWS r6id.8xlarge, 256GiB RAM, 32×Intel Xeon Platinum 8375C, 2 read threads, 2 write threads |
216,222 USD | 6.76 ops/s | ||
| 30 | AWS r6id.8xlarge, 256GiB RAM, 32×Intel Xeon Platinum 8375C, 4 read threads, 4 write threads |
216,222 USD | 12.16 ops/s |
SNB Business Intelligence workload – Audited results
| Benchmark setup | SF | Hardware | System cost | Performance metrics | Documents |
|---|---|---|---|---|---|
|
1000 | Dell PowerEdge 6625 with AMD EPYC 9354 |
1,353,315 USD |
|
LDBC-certified auditors
SNB Interactive audits can be commissioned from the following LDBC-certified auditors:
- Márton Búr
- Arnau Prat-Pérez
- David Püroja
SNB BI audits can be commissioned from the following LDBC-certified auditors:
- Fabian Murariu (Pometry Ltd.)
- David Püroja
For auditing requests, please reach out at info@ldbcouncil.org.
Audits can only be commissioned by LDBC member companies by contracting any of the LDBC-certified auditors.
Additionally, there is a 2,000 GBP auditing fee to be paid for the LDBC for non-sponsor company members. Sponsor companies are exempt from this.
Use of audited results
Fair use policies
The LDBC Social Network Benchmark is subject to the LDBC Fair Use Policies.
Retrospective review of publications related to LDBC benchmark standards
- Review of 2019 preprint (sponsored by TigerGraph)
- Review of 2020 whitepaper (authored by Oracle)
- Review of the keynote at NODES 2021 (Neo4j Online Developer Expo and Summit)