README
¶
FSC Node GRPC View API benchmarks
This package exercises the GRPC View API implementation platform/view/services/view/grpc.
It contains a gRPC client/server benchmark designed to measure throughput (TPS) and tail latency (p99).
The benchmark supports multiple workloads views ranging from no-op calls to CPU-intensive cryptographic operations (e.g., ECDSA signing), and enables systematic comparison, and deployment modes (local vs remote).
The benchmark workloads views are defined in integration/benchmark/views package.
Benchmarks are executed with configurable warmup and measurement phases to ensure that results reflect steady-state behavior rather than cold-start effects.
On the client side, load is generated using a configurable number of worker goroutines mapped onto a configurable number of pre-created gRPC connections. Workers continuously issue RPCs until a deadline is reached, recording per-request latency into histograms after the warmup period. Benchmark execution is deadline-driven rather than iteration-driven, ensuring consistent measurement intervals across runs.
Note that the TLS enabled by default but the client sets InsecureSkipVerify=true.
This is an example output of the benchmark showing the number of executed request, the TPS, and the p5, p95, p99 latencies.
BenchmarkAPIGRPCRemote/w=noop/nc=2-8 130048 13005 TPS 597368 ns/op (p5) 753319 ns/op (p95) 863278 ns/op (p99)
BenchmarkAPIGRPCRemote/w=noop/nc=2-16 236702 23670 TPS 654140 ns/op (p5) 847374 ns/op (p95) 1396724 ns/op (p99)
plot_node.py is a script that parses raw benchmark output and generate grouped throughput plots across workloads and connection counts.
Usage
The server can be started as follows:
go run ./server/main.go
The server generates the crypto material folder in out/testdata.
The benchmarks can be run as follows:
# get the testdata folder from the server and place it next to our benchmark client
rsync -av --exclude='.git' --exclude='env/' user@myserver.com:/home/user/fabric-smart-client/integration/benchmark/node/remote /home/user/fabric-smart-client/integration/benchmark/node/remote
# update the connection file and change server address from 127.0.0.1:8099 to your server address, for example, `myserver.com:8099`
vim /home/user/fabric-smart-client/integration/benchmark/node/remote/out/testdata/fsc/nodes/test-node.0/client-config.yaml
go run ./client/main.go -benchtime=10s -count=5 -workloads=noop,cpu,sign -cpu=1,8,16,32,48,64,96,128,256,376,512,640,768,1024 -numConn=1,2,4,8 2>&1 | tee results.txt
Directories