Episode 421: Doug Fawley on gRPC

Benchmark Analysis

In examining GRPC's performance benchmarks, Doug Fawley highlights several key insights. He explains that typical benchmarks focus on micro-level performance, such as latency and throughput. These benchmarks test how quickly a single request can be processed through a GRPC client and server application stack, and how many bytes can be pushed through the system under various conditions 1.

Our typical benchmarks are a little bit more micro benchmark focused. We have benchmarks that are intended to stress the latency of a single operation.

--- Doug Fawley

Fawley notes that these tests are crucial for understanding the efficiency and capacity of GRPC in handling different loads 2 3.

   

QPS Metrics

The discussion on queries per second (QPS) metrics reveals significant insights into GRPC's system capabilities. Doug Fawley describes a QPS benchmark that measures how many raw messages per second can be handled, emphasizing its importance in evaluating system performance 1. He shares that in a Google Compute Engine environment, the Java client can handle approximately 750,000 RPCs per second, showcasing GRPC's robust performance 4.

I see QPS benchmark, which is queries per second that are able to be handled by a single vm. This is running in GCE, Google Compute engine, and it looks like I'm seeing about 750,000 for the Java client.

--- Doug Fawley

These metrics are essential for developers to understand the potential and limitations of GRPC in high-demand scenarios.