Skip to content

Latest commit

 

History

History
127 lines (111 loc) · 2.79 KB

BENCHMARK.md

File metadata and controls

127 lines (111 loc) · 2.79 KB
Raw

Benchmark Results

You will find comparison results for warden with request. Used nock to simulate HTTP server and custom test runner to measure results. You can run same tests by yourself

TS_NODE_FILES=true node -r ts-node/register  benchmarks/holder.ts

Tests

Raw

All modules are disabled on warden.

[
    {
        "name": "Request Module",
        "stats": {
            "runs": 10000,
            "success": 10000,
            "errors": 0,
            "duration": 16827.367216
        }
    },
    {
        "name": "Warden",
        "stats": {
            "runs": 10000,
            "success": 10000,
            "errors": 0,
            "duration": 16868.094933
        }
    }
]

Warden puts 0.24% overhead on request module. Almost nothing.

Holder

Only holder module enabled. Running on 500 parallel requests with 1/100 chance of same request. This means only 5 same concurrent requests.

[
    {
        "name": "Request Module",
        "stats": {
            "runs": 100000,
            "success": 100000,
            "errors": 0,
            "duration": 17612.177992
        }
    },
    {
        "name": "Warden",
        "stats": {
            "runs": 100000,
            "success": 100000,
            "errors": 0,
            "duration": 17501.875109
        }
    }
]

Warden wins this one clearly. Results will be better for warden when transferring over real network.

Cache

Cache module active. Running with 100 parallelism and 1000 different requests.

[
    {
        "name": "Request Module",
        "stats": {
            "runs": 1000,
            "success": 1000,
            "errors": 0,
            "duration": 1702.591433
        }
    },
    {
        "name": "Warden",
        "stats": {
            "runs": 1000,
            "success": 1000,
            "errors": 0,
            "duration": 22.202227
        }
    }
]

This one is no-brainer. Cached one will be faster all the time. But cache implementation on Warden is amazingly fast.

Retry

Retry module active. Nock is configured to simulate a production class server with >0.99 success rate. Running tests 100 parallelism and 1 retry amount.

[
    {
        "name": "Request Module",
        "stats": {
            "runs": 5000,
            "success": 4956,
            "errors": 44,
            "duration": 4216.685647
        }
    },
    {
        "name": "Warden",
        "stats": {
            "runs": 5000,
            "success": 5000,
            "errors": 0,
            "duration": 4222.504931
        }
    }
]

request module is barely faster but it resulted with 44 errors. Meanwhile Warden has full success rate.