README_benchmark.md

BENCHMARK

Using jmh for benchmarking the leader node

Start leader

make run-leader

Benchmark

./gradlew jmh

Using simple code for benchmarking the leader node

We open only one grpc connection from the benchmark tool to the leader node. (In real-world, one client-app must open only one connection to the leader node.)

Start leader

make run-leader

Start benchmark tool

make run-benchmark-tool

Benchmark deposit

curl --location --request POST 'http://localhost:8900/api/balance-benchmark/benchmark/500000/64'

Result of sending 500,000 write requests with 64 grpc connections to cluster:

Using Ghz for benchmarking the cluster nodes (leader, follower)

We open many grpc connections to the leader node.

• Deposit

ghz \
--insecure \
--skipFirst 100000 \
--proto ./bank-libs/bank-cluster-proto/src/main/proto/balance.proto \
--call gc.garcol.bank.proto.BalanceCommandService/sendCommand \
-d '{
    "depositCommand": {
        "id": 1,
        "amount": 1,
        "correlationId": "random-uuid"
    }
}' \
-c 200 -n 600000 \
127.0.0.1:9500

Benchmark client-nodes using autocannon

autocannon \
--warmup [ -c 32 -d 20 ] \
-c 32 -d 30 \
-m POST \
-H 'Content-Type: application/json' \
-b '{"id":"1","amount":"1"}' \
http://localhost:8900/api/v1/balance/command/deposit

┌─────────┬──────┬────────┬────────┬────────┬───────────┬───────┬─────────┐
│ Stat    │ 2.5% │ 50%    │ 97.5%  │ 99%    │ Avg       │ Stdev │ Max     │
├─────────┼──────┼────────┼────────┼────────┼───────────┼───────┼─────────┤
│ Latency │ 8 ms │ 140 ms │ 326 ms │ 359 ms │ 147.04 ms │ 99 ms │ 1298 ms │
└─────────┴──────┴────────┴────────┴────────┴───────────┴───────┴─────────┘
┌───────────┬─────────┬─────────┬────────┬─────────┬───────────┬────────┬─────────┐
│ Stat      │ 1%      │ 2.5%    │ 50%    │ 97.5%   │ Avg       │ Stdev  │ Min     │
├───────────┼─────────┼─────────┼────────┼─────────┼───────────┼────────┼─────────┤
│ Req/Sec   │ 9,079   │ 9,079   │ 10,287 │ 12,375  │ 10,478.94 │ 848.02 │ 9,072   │
├───────────┼─────────┼─────────┼────────┼─────────┼───────────┼────────┼─────────┤
│ Bytes/Sec │ 1.94 MB │ 1.94 MB │ 2.2 MB │ 2.65 MB │ 2.24 MB   │ 181 kB │ 1.94 MB │
└───────────┴─────────┴─────────┴────────┴─────────┴───────────┴────────┴─────────┘

Req/Bytes counts sampled once per second.
# of samples: 30

317k requests in 30.03s, 67.2 MB read

The results of the test indicate that the client-app is capable of handling 12,000 write requests per second. Given the cluster's maximum capacity of 200,000 write requests per second, we can achieve this throughput by scaling the client-app to 16 instances. This will ensure that we fully utilize the cluster's capacity.