Very high performance web application daemon
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


Very high performance web application daemon

Copyright 2014 - 2017 by Jarle (jgaa) Aase. All rights reserved. This project is licensed under the GNU version 3 license. Please see the LICENSE file.

You can contact the developer at


A few years ago, a company issued an annual report about the popularity of web (HTTP) servers. Each year, Microsoft IIS achieved ridicules high scores. At that time I worked for a huge (top 10) US software company as a senior C++ guy, and in a lunch break we discussed this madness. Everyone agreed that IIS cannot compete in the web server space. It's all about Apache and nginx. Sure - there are lots of companies running IIS to host their web pages, but they are getting fewer as Internet becomes a more hostile place, and front facing servers require professional attention.

So why could this be? I looked at the report, and noticed that the popularity was based on number of distinct sites (host names) - not IP numbers served, or number of distinct page hits. I suggested that these absurd numbers may be deliberately manipulated by Microsoft, if they hosted millions of idle sites in their data-centers, using free cloud servers to simply answer on request from the web-spiders that create such reports.

At that time there was about 1 billion distinct web sites, according to this report. As a joke, I suggested that I could write a HTTP server by myself and host another 1 billion web sites, and make my web server the most popular on the Internet in a week!

It was actually a fun idea, and as anyone who have worked in a large software company knows, progress there is s l o w . . .

So I started to code my 1 billion site web server on my spare time, finally making some progress. I wrote an embedded DNS server in 16 hours over 3 days. I spent the same 3 days, at work, getting a small, insignificant bug-fix trough code review and submit process (where the tests failed and failed and failed because of errors in the testing infrastructure).

About a month later, I ran 1 billion distinct web sites on my laptop, (ThinkPad w520 with 16 GB ram and Debian Linux). The performance tests (querying from another machine) showed that the server handled 170.000 page requests per second. That maxed out the bandwidth on the 1GB network interface.

vUbercool can really serve 1 billion websites with distinct state (32 bits state), a few real websites with static content and a number of sites using C++ generated content from different content generators. To serve 1 billion host names (from the same domain name), it has an embedded DNS server.

The server is written in C++11, using boost::asio for IO.


I bought the domain name, but never found an internet provider that could deliver the bandwidth I needed to host 50% of the Internet on my laptop at home.

However, I still like the idea, so I plan to keep the code working, and use my own wfde library as much as possible. That allows me to tune wfde for performance, and use vUbercool as a showcase and benchmark tool - and may be the foundation of a real web application server in C++ in the future (my old work-horse WarCMS is long overdue for replacement).

Anyway, the code for the original project is here. Have fun.

How to test

Build the project

~$ mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
sudo make install

If you really want to give it a go, you have to change the dns server to the local ip, as the built-in DNS server will resolve the hostnames. The example configuration files use, but you can use anything you want.

You can do that by changing /etc/resolv.conf

~$ cat /etc/resolv.conf

Then, give the server permissions to open ports below 1024 (needed for dns).

~$ sudo setcap 'cap_net_bind_service=+ep' /usr/local/bin/vubercoold

Modify the configuration-files in conf to suit your needs.

Start the server for the first time

~/src/vUbercool$ vubercoold --http-config conf/http.conf --dns-config conf/dns.conf -C DEBUG --recreate-mmap-file true

Note that the directory for the memory map file (specified in http.conf) must exist and be writable for the user running the demon.

Have fun.