0016 0046 0003 0007 0054 nxweb – Fast and Lightweight Web Server for Applications Written in C & Python 0008 0020 0028 0003 0004 0027 0003 0003 0040 0072 0097 01B0 0002 0005 0064 041E 0005 0004 0079
0034
000B 000D 0051

0054 nxweb – Fast and Lightweight Web Server for Applications Written in C & Python 0017

0013

000F 127C

Rationale

Sometimes web applications need small and fast components that are best written in C.

Example: ad banner rotation engine. It gets invoked many times on every page of your site producing little HTML snippets based on predefined configuration. How would you implement it?

CGI is not an option as it gets loaded and unloaded on every request. Writing a module for your main web server such as Apache httpd or nginx gives best performance but server's API isn't very friendly (especially when dealing with shared memory, etc.). What we need is sort of servlet container for C (just like Java has).

Before writing NXWEB I evaluated a number of existing light/embeddable web servers (mongoose, microhttpd, libevent, G-WAN) each one having their own drawbacks (see notes to benchmarks).

What NXWEB offers:

  • good (if not best) performance; see benchmarks
  • can serve thousands concurrent requests
  • small memory footprint
  • event-driven & multi-threaded model designed to scale
  • exceptionally light code base
  • simple C API
  • decent HTTP protocol handling
  • keep-alive connections
  • chunked requests and responses
  • SSL support (via GNUTLS)
  • HTTP proxy (with keep-alive connection pooling)
  • file cache for proxied content and custom handlers' output
  • cached content can be served when backend is unavailable
  • non-blocking sendfile support (with configurable small file memory cache)
  • cacheable gzip content encoding
  • cacheable image thumbnails with watermarks (via ImageMagick)
  • basic server-side includes (SSI)
  • templating engine with page inheritance
  • subrequests
  • integrated Python interpreter and WSGI-server
  • modular design for developers
  • can be run as daemon; relaunches itself on error
  • open source

Limitations:

  • only tested on Linux

Architecture

Main thread binds TCP port, accepts connections and distributes them among network threads.

Network threads work in non-blocking fashion (using epoll) handling HTTP protocol exchange. There should be no need in using more network threads than the number of CPU cores you have, as every thread is very efficient and can easily handle thousands of concurrent connections. Numbers of network threads is automatically configured but could be limited by #define NXWEB_MAX_NET_THREADS directive.

After receiving complete HTTP request network thread finds and invokes application's URI handler. The invocation could happen within network thread (should only be used for quick non-blocking handlers) or within worker thread (should be used for slower or blocking handlers).

Worker threads are organized in pool. Each thread takes new job from queue, invokes URI handler, signals network thread after handler completes the job. Number of worker threads is dynamically configured depending on the load with limits specified by #define directives.

API

Read articles on this site + see source code. hello.c is an example of user module. nxweb.h contains function and struct definitions.

Benchmarks

See NXWEB benchmarks here.

Building from source

  1. Download archive from here
  2. Follow INSTALL file notes

Issues

Please report issues via issue tracker.

Discussions & Announcements

nxweb
Visit this group

nxweb-ru (in Russian)
Visit this group

0022
0045
0049
0004 01FA 0002 0002 0045 0036 0001 0002 0004 0001 0011 0