Polygon Benchmarks

Benchmark Date: 08/24/2023

Providers included

  • Nodies

  • Blast API

  • Ankr

  • Quicknode

  • Alchemy

RPC Method Tested

eth_getBlockByNumber

Flood Command

flood eth_getBlockByNumber <endpoint1> <endpoint2> <endpoint3> <endpoint4> <endpoint5> --rates 10 100 1000 10000 --duration 30

Benchmark Summary

When benchmarking the Polygon mainnet endpoints across various request loads, the performance of the endpoints presented noticeable differences:

Nodies:

  • P50 Latency: Showcased unparalleled performance across all loads, maintaining sub-7ms latency even at the highest request rate of 10,000 rps.

  • P90 Latency: Impressively consistent, with its latency remaining under 9ms up to 1,000 rps, and surging modestly to 411ms at 10,000 rps.

  • P99 Latency: Demonstrated consistent efficiency, never surpassing 1.75 seconds, even at the most demanding load of 10,000 rps.

Other Endpoints:

  • P50 Latency: Performance varied with BlastAPI showing promise in the lower loads. QuickNode, Alchemy, and Ankr faced challenges, especially as loads increased.

  • P90 Latency: All endpoints, excluding Nodies, showed increased latencies as loads grew. Notably, QuickNode's latency neared an alarming 40.8 seconds at 10,000 rps.

  • P99 Latency: The discrepancies widened, with most services, especially QuickNode, facing challenges at the highest loads, reaching latencies of 48.3 seconds and 16 seconds, respectively.

In conclusion, Nodies distinctly outperformed the rest, consistently delivering optimal latencies across the board. The other endpoints varied in their responses, with some like BlastAPI showing potential, while others, particularly QuickNode at higher loads, experienced significant latencies. Overall, for consistent performance, Nodies stood out as the prime choice among the tested endpoints.

Latency Benchmarks

Metrics: Latency percentiles (P50, P90, P99) against load rates (10, 100, 1000, and 10000 rps). P50 Latency (in ms, lower is better):

P90 Latency (in ms, lower is better):

P99 Latency (in ms. lower is better):

Last updated