Performance

Early tests are promising, but there are many variants on how Proksi performs under real load.

There are also some optimizations that can be done to improve performance in the long term, though the focus is on making it feature complete first.

Wrk benchmark

An sample run from the wrk benchmark on the simple /ping endpoint shows the following results:

# Apple M1 Pro, 16GB
# Memory Usage: 15.2MB, CPU Usage: 13%, 1 worker thread
# at 2024-06-13
# Wrk > 50 connections, 4 threads, 30s duration
wrk -c 50 -t 4 -d 30s http://127.0.0.1/ping

Running 30s test @ http://127.0.0.1/ping
  4 threads and 50 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   376.73us  124.78us   7.83ms   91.64%
    Req/Sec    31.76k     3.03k   34.29k    92.44%
  Latency Distribution
     50%  373.00us
     75%  404.00us
     90%  442.00us
     99%  675.00us
  3803987 requests in 30.10s, 467.98MB read
Requests/sec: 126377.09
Transfer/sec:     15.55MB

CDN performance

Using the following configuration we can see that Proksi holds its own in regards to performance.

cache {
  enabled = true
  path = "/tmp"
  cache_type = "disk"
}

1 minute test using 4000 concurrent clients.