Concurrent Web Server in C

This project implements a simple concurrent web server in C, designed to handle multiple requests simultaneously using multithreading. The server is capable of serving static files and executing CGI scripts.

Design Overview

Usage

$ ./wserver -d <directory> -p <port> -t <threads> -s <schedule_policy> -b <max_connections>

Test the multithreading with httperf and spin.cgi

We use httperf to simulate multiple clients making requests to the server. Since the spin.cgi script is designed to take a 1 second pause, we can use it to test the server's ability to handle concurrent requests (while the thread issuing the I/O request suspends for waiting the I/O to complete, other threads can run, thus enabling the server to make progress). We test the following server configurations:

1. 100 threads with max 1000 connections
2. 10 threads with max 1000 connections

We expect the config 1 to handle the requests efficiently, while config 2 should struggle due to the limited number of threads.

Runtime Specifications

• CPU: 2.3 GHz Quad-Core Intel Core i5
• Memory: 8 GB 2133 MHz LPDDR3
• OS: macOS 15.3.1

Config 1: 100 Threads, 1000 Max Connections

$ ./wserver -d . -p 8000 -t 50 -b 1000
$ httperf --server 0.0.0.0 --port 8000 --uri "/spin.cgi?1" --num-conns 1000 --rate 100
httperf: warning: open file limit > FD_SETSIZE; limiting max. # of open files to FD_SETSIZE
Maximum connect burst length: 1

Total: connections 1000 requests 1000 replies 1000 test-duration 11.044 s

Connection rate: 90.5 conn/s (11.0 ms/conn, <=106 concurrent connections)
Connection time [ms]: min 1003.7 avg 1028.8 max 1067.6 median 1028.5 stddev 15.6
Connection time [ms]: connect 0.0
Connection length [replies/conn]: 1.000

Request rate: 90.5 req/s (11.0 ms/req)
Request size [B]: 70.0

Reply rate [replies/s]: min 79.8 avg 89.5 max 99.2 stddev 13.7 (2 samples)
Reply time [ms]: response 24.4 transfer 1004.4
Reply size [B]: header 90.0 content 127.0 footer 0.0 (total 217.0)
Reply status: 1xx=0 2xx=1000 3xx=0 4xx=0 5xx=0

CPU time [s]: user 1.64 system 9.13 (user 14.9% system 82.7% total 97.5%)
Net I/O: 25.4 KB/s (0.2*10^6 bps)

Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0

Config 2: 10 Threads, 1000 Max Connections

$ ./wserver -d . -p 8000 -t 10 -b 1000
$ httperf --server 0.0.0.0 --port 8000 --uri "/spin.cgi?1" --num-conns 1000 --rate 100
httperf --client=0/1 --server=0.0.0.0 --port=8000 --uri=/spin.cgi?1 --rate=100 --send-buffer=4096 --recv-buffer=16384 --num-conns=1000 --num-calls=1
httperf: warning: open file limit > FD_SETSIZE; limiting max. # of open files to FD_SETSIZE
Maximum connect burst length: 1

Total: connections 1000 requests 1000 replies 1000 test-duration 100.530 s

Connection rate: 9.9 conn/s (100.5 ms/conn, <=910 concurrent connections)
Connection time [ms]: min 1003.7 avg 45783.3 max 90584.6 median 45346.5 stddev 26125.2
Connection time [ms]: connect 0.0
Connection length [replies/conn]: 1.000

Request rate: 9.9 req/s (100.5 ms/req)
Request size [B]: 70.0

Reply rate [replies/s]: min 8.0 avg 9.9 max 10.0 stddev 0.4 (20 samples)
Reply time [ms]: response 44779.5 transfer 1003.8
Reply size [B]: header 90.0 content 127.0 footer 0.0 (total 217.0)
Reply status: 1xx=0 2xx=1000 3xx=0 4xx=0 5xx=0

CPU time [s]: user 6.38 system 93.73 (user 6.3% system 93.2% total 99.6%)
Net I/O: 2.8 KB/s (0.0*10^6 bps)

Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0

Analysis

The config 1 is 10 times faster than config 2.

Simple Test on Shortest File First Schedule Policy

We added sleep(10) in line 117 to allow simply manually send some HTTP requsts to the server for getting files with various size:

• random_500: 136175 bytes
• random_100: 27977 bytes

Request Worker Code Snippest:

   115	  } else if (SCHE_POLY == SCHE_POLY_SFF){
   116	      while (1) {
   117	        // sleep(10); // for testing purpose which allow buffering the requests in the min heap
   118	        HTTPRequest *req = NULL;
   119	        int conn_fd = get_conn_fd(&req);
   120	        assert(conn_fd != 0);
   121	        assert(req != NULL);
   122	        printf("Serving uri:%s\n", req->uri);
   123	        request_handle_without_parse(conn_fd, req);
   124	        close_or_die(conn_fd);
   125	        free(req);
   126	      }
   127	  }
   128	}

The below lines are the output of web server that running with shortest file first policy. We can see that the server first receive the request of getting file /random_500 then the request of getting file /random_100. As we expected, the server first serve /random_100 and then serve /random_500.

❯ ./wserver -p 8080 -d "./static" -t 1 -b 10 -s 1
method:GET uri:/random_500 version:HTTP/1.1
method:GET uri:/random_100 version:HTTP/1.1
Serving uri:/random_100
Serving uri:/random_500