Hi.
I have been testing (again) MHD and Nginx with a static content and noticed
MHD is around two times slower than Nginx, but I'm not sure if the slowness
is related to MHD or something I'm doing wrong.
*Test built with MHD*
#include <stdio.h>
#include <memory.h>
#include <microhttpd.h>
#define PAGE
\
"<!DOCTYPE html>\n\
<html lang=\"en\">\n\
<head>\n\
<meta charset=\"UTF-8\" />\n\
<meta name=\"viewport\" content=\"width=device-width,
initial-scale=1.0\" />\n\
<title>Hello world benchmark</title>\n\
</head>\n\
<body>\n\
This is a static content to check the performance of the following
HTTP\n\
servers:\n\
<ul>\n\
<li>MHD</li>\n\
<li>nginx</li>\n\
</ul>\n\
</body>\n\
</html>"
static enum MHD_Result ahc_echo(void *cls, struct MHD_Connection *con,
const char *url, const char *method,
const char *version, const char
*upload_data,
size_t *upload_data_size, void **ptr) {
struct MHD_Response *res;
enum MHD_Result ret;
if ((void *)1 != *ptr) {
*ptr = (void *)1;
return MHD_YES;
}
*ptr = NULL;
res = MHD_create_response_from_buffer(strlen(PAGE), PAGE,
MHD_RESPMEM_PERSISTENT);
ret = MHD_queue_response(con, MHD_HTTP_OK, res);
MHD_destroy_response(res);
return ret;
}
int main() {
struct MHD_Daemon *d;
d = MHD_start_daemon(
MHD_USE_EPOLL_INTERNAL_THREAD | MHD_SUPPRESS_DATE_NO_CLOCK |
MHD_USE_EPOLL_TURBO,
8080, NULL, NULL, &ahc_echo, NULL, MHD_OPTION_CONNECTION_TIMEOUT,
(unsigned int)120, MHD_OPTION_THREAD_POOL_SIZE,
(unsigned int)sysconf(_SC_NPROCESSORS_ONLN),
MHD_OPTION_CONNECTION_LIMIT,
(unsigned int)10000, MHD_OPTION_END);
getchar();
MHD_stop_daemon(d);
return 0;
}
*Nginx stuff*
Version: 1.18.0
Config:
$ cat /etc/nginx/nginx.conf
worker_processes auto;
worker_cpu_affinity auto;
events {
worker_connections 10000;
}
http {
access_log off;
keepalive_timeout 65;
server {
listen 8080 default_server;
...
$ cat /usr/share/nginx/html/index.html
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Hello world benchmark</title>
</head>
<body>
This is a static content to check the performance of the following HTTP
servers:
<ul>
<li>MHD</li>
<li>nginx</li>
</ul>
</body>
</html>
*Environment*
$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 43 bits physical, 48 bits virtual
CPU(s): 8
On-line CPU(s) list: 0-7
Thread(s) per core: 2
Core(s) per socket: 4
Socket(s): 1
NUMA node(s): 1
Vendor ID: AuthenticAMD
CPU family: 23
Model: 24
Model name: AMD Ryzen 7 3700U with Radeon Vega Mobile
Gfx
Stepping: 1
Frequency boost: enabled
CPU MHz: 2100.611
CPU max MHz: 2300.0000
CPU min MHz: 1400.0000
BogoMIPS: 4591.70
Virtualization: AMD-V
L1d cache: 128 KiB
L1i cache: 256 KiB
L2 cache: 2 MiB
L3 cache: 4 MiB
NUMA node0 CPU(s): 0-7
$ cat /proc/version
Linux version 5.6.14-300.fc32.x86_64 (
mockbu...@bkernel03.phx2.fedoraproject.org) (gcc version 10.1.1 20200507
(Red Hat 10.1.1-1) (GCC)) #1 SMP Wed May 20 20:47:32 UTC 2020
*Finally, the tests using wrk!*
*wrk* results (avg, after three intervaled tests) for Nginx:
Running 10s test @ http://127.0.0.1:8080/
10 threads and 1000 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 7.71ms 3.61ms 72.10ms 74.14%
Req/Sec 12.37k 1.46k 26.09k 82.34%
Latency Distribution
50% 6.96ms
75% 9.29ms
90% 12.65ms
99% 18.26ms
1228942 requests in 10.09s, 717.21MB read
Requests/sec: 121831.57
Transfer/sec: 71.10MB
*wrk* results (avg) for MHD
Running 10s test @ http://127.0.0.1:8080/
10 threads and 1000 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 12.18ms 7.47ms 59.49ms 71.56%
Req/Sec 7.50k 1.54k 16.48k 68.81%
Latency Distribution
50% 10.72ms
75% 16.23ms
90% 22.37ms
99% 35.05ms
745561 requests in 10.09s, 310.72MB read
Requests/sec: 73908.50
Transfer/sec: 30.80MB
Those tests were done in localhost just to illustration, but I did the same
tests at an external server remotely (with some limits due to internet
bandwidth) providing a larger content (around 150 kB) and got slower
results with MHD.
I don't know if there is any specific configuration to increase the MHD
speed again. BTW I would appreciate it and retest to get better results.
TIA for any help! 👍
--
Silvio Clécio
