PlanetScale Hits 118 Million Queries Per Second in Neki Stress Test
A new platform preview demonstrates linear scalability across 512 shards, handling massive read-only workloads.
PlanetScale has successfully pushed its new Neki platform preview to a massive scale, sustaining over 118 million queries per second (QPS). The benchmark serves as a technical validation of the system's ability to scale horizontally as demand increases.
According to PlanetScale, the stress test sustained 118,538,803 QPS for a duration of 16 minutes, reaching a peak of 118,747,267 QPS. To achieve these figures, the company deployed an infrastructure consisting of 512 shards, each featuring one Postgres primary on r8g.16xlarge instances, supported by 480 Neki routers running on 8xlarge instances. The workload was specifically designed as a read-only test, utilizing single-shard point selects by primary key without the use of writes, joins, or cross-shard queries.
The Scale of the Benchmark
The test operated on a massive dataset totaling 1.22 PiB. This volume of data generated significant infrastructure pressure, resulting in over 2 Tb per second of network traffic and 15.8 million read IOPS. Despite the extreme throughput, PlanetScale reported stable latency metrics, with p99 latency measured at 6.06ms at the router level and 13.95ms at the client level.
Why Linear Scalability Matters
The primary objective of the benchmark was to test Neki's horizontal scaling capabilities. By incrementally increasing the number of shards from 5 to 50 and eventually to 512, PlanetScale demonstrated linear scalability. This means that throughput increased proportionally with the addition of hardware, suggesting that users can theoretically scale their database performance predictably by adding more resources.
For the broader industry, these results highlight the potential for Neki to handle high-throughput database workloads that would overwhelm traditional architectures. While the test focused on the simplest possible query type—the point select—the sheer volume of data and network traffic handled proves the platform's underlying stability at an extreme scale.
Looking Ahead
Neki remains in platform preview as PlanetScale continues to refine the offering. While the current benchmark confirms the platform's ceiling for read-heavy, single-shard operations, the next phase of evaluation will likely involve more complex workloads. It remains to be seen how the system maintains this linear efficiency when introducing writes, complex joins, or the cross-shard queries that were omitted from this specific stress test.