Overview

A collection of networking stuff for making network (client/server) application.

compilation and run

The program has been tested under the Linux 5.15.153.1-microsoft-standard-WSL2

$ mkdir build && cd build && cmake ..
$ make        # complation
$ ./example/http_srv  # run server

dependencies

• boost

sudo apt install libboost-dev

brew instal boost

Features

• http/web socket server
• support transfer-encoding chunk

http

I have done couple of experimental results, libevent-http_parser, libuv-http_parser, asio-http_parser.

All of them are performant, however I have personally decided to select asio as IO event and http-parser for a http server. And its performance can achieve 220,000 request per seconds on ubuntu-22.04/intel i5-1135G7 @ 2.40GHz

http server example

you can quick start creating a http server by the below source code.

int main(int argc, char ** argv)
{
    std::string addr = "127.0.0.1";
    uint16_t port    = 12345;
    http::start_server(port, [](http::response res) {
        res.body() = "[start_server_ng] hello";
        res.send();
    });
}

websocket echo server

int main(int argc, char * args[])
{
    if (argc != 3)
    {
        std::cerr << "\nUsage: " << args[0] << " address port\n" << "Example: \n" << args[0] << " 0.0.0.0 12345" << std::endl;
        return -1;
    }

    std::string addr(args[1]);
    uint16_t port   = static_cast<uint16_t>(std::atoi(args[2]));
    auto ws_server_ = ws::make_server(
        port,
        [](ws::message msg) {
            msg.data_out() << "Echo: " << msg.data();
            msg.send();
        },
        [](beast::error_code ec) {});

    io_context::ioc.run();
}

Performance

There are several criterias for measuring the performance of a http server.

And it varies from the hardward and OS. THe performance is measured on environment:

• OS: Linux 5.15.153.1-microsoft-standard-WSL2
• Hardware: 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz

The ab tool is used to measure the performance.

Request per second

It is tested by 50 concurrency active connection and 100,000 request in total.

$ ab -k -c 50 -n 100000 127.0.0.1:12345/route_01

http server	Request per second	Remark
av_connect	~220,000 rps	release-0.0.4
nodejs	~12,000 rps	v12.22.9
asiohttp	~11,000 rps	3.10.6
flask	~697 rps	3.0.3

Comparing with other http framework, which is found at here

its application

• OpenAI API server for serving LLM inference

Reference

• HTTP/1.0 https://datatracker.ietf.org/doc/html/rfc1945
• HTTP/1.1 https://datatracker.ietf.org/doc/html/rfc2616