This topic provides an overview of RisingWave’s ad-hoc query serving performance characteristics (latency and throughput) based on the industry-standard sysbench benchmark.
sbtest
was created in RisingWave with the following schema:
sbtest
table. Two data sizes were tested: 1 million rows (approximately 186MB) and 10 million rows (approximately 1.86GB).
oltp_point_select
:
select_random_points
:
select_random_ranges
:
Latency (ms) | Min | Avg | P95 | P99 | Max | Throughput (QPS) |
---|---|---|---|---|---|---|
oltp_point_select | 0.40 | 5.05 | 9.22 | 15.27 | 131.75 | 25,335 |
select_random_points | 1.27 | 15.11 | 21.89 | 31.94 | 131.87 | 8,467 |
select_random_ranges | 1.92 | 13.98 | 20.00 | 35.59 | 233.90 | 9,156 |
Latency (ms) | Min | Avg | P95 | P99 | Max | Throughput (QPS) |
---|---|---|---|---|---|---|
oltp_point_select | 0.40 | 4.96 | 8.90 | 18.28 | 233.90 | 25,814 |
select_random_points | 0.84 | 16.40 | 24.38 | 32.53 | 175.82 | 8,203 |
select_random_ranges | 0.86 | 13.84 | 19.65 | 38.94 | 142.34 | 9,247 |
oltp_point_select
workload, the frontend node becomes the primary bottleneck due to the lightweight nature of the operations. For select_random_points
and select_random_ranges
, both the compute and frontend nodes experience CPU saturation.