Sysbench benchmark results
These are Sysbench results for MongoDB Enterprise across compute flavors, published so you can size a cluster before provisioning it.
These figures are reference data from a synthetic workload, not a performance guarantee. Your throughput will differ. Test with your own workload before committing to a production configuration.
What was measured
The benchmark evaluates OLTP read/write performance, shows how throughput scales with CPU and memory, and gives a basis for choosing an instance size.
| Metric | Meaning |
|---|---|
| Sysbench Read | Total read queries executed during the test |
| Sysbench Write | Total write queries executed during the test |
| QPS | Average queries processed per second |
| TPS | Average completed transactions per second |
Higher QPS and TPS mean higher throughput under the tested workload.
Test environment
| Parameter | Value |
|---|---|
| Benchmark tool | Sysbench (OLTP Read/Write) |
| Number of documents | 1,000,000 |
| Number of operations | 1,000,000 |
| Workload type | Read/Write |
| Thread count | Varies per flavor — see results |
| MongoDB version | MongoDB Enterprise v8.0.20-ent |
| IOPS | 6,000 |
Results
| Flavor (vCPU/RAM) | Threads | Sysbench Read | Sysbench Write | QPS | TPS |
|---|---|---|---|---|---|
| 2C4G | 8 | 500,310 | 499,690 | 4,083.95 | 4,083.95 |
| 2C8G | 8 | 500,652 | 499,348 | 5,120.56 | 5,120.56 |
| 4C8G | 16 | 499,868 | 500,132 | 9,007.79 | 9,007.79 |
| 8C16G | 32 | 501,398 | 498,602 | 17,666.60 | 17,666.66 |
| 8C32G | 64 | 500,420 | 499,529 | 21,993.49 | 21,993.49 |
| 16C32G | 64 | 499,522 | 500,478 | 31,626.55 | 31,626.55 |
| 16C64G | 128 | 500,086 | 499,914 | 34,621.24 | 34,621.24 |
Reading the numbers
Throughput rises with CPU and memory, but not linearly. Going from 2C4G to 2C8G (doubling RAM alone) adds about 25% QPS, while 4C8G to 8C16G roughly doubles it. Gains flatten at the higher configurations as other limits take over.
Your actual performance depends on:
- Application workload characteristics
- Schema design and indexing
- Read/write ratio
- Storage and network configuration
Note the test ran at 6,000 IOPS. At a lower storage policy, I/O becomes the bottleneck and additional vCPU will not deliver these numbers.
Next steps
- Create your first database, where you pick the flavor and storage policy
- Change database resource configurations to resize an existing cluster