Preface
========

	It took me waaaay more than 2 hours to complete this task due to boilerplate
code necessary to "bootstrap" the whole thing. In my practice I try to write
reasonable C++ wrappers around plain C code to minimize human mistakes related
to memory handling or ownership control.
	In my opinion, 2 hours for such task is unreasonably small amount of time
to "aim for perfection" at least taking "perfection" as I personally understand
it based on my SW development practice.

Building
========
	Project was developed on Ubuntu 14.04 with Gnu autotools.
	Prior to invoking autotools make sure you have curl dev library installed:

		$ sudo apt-get install libcurl4-openssl-dev

	Then invoke autotools as usual:

		$ autoreconf -f -i
		$ ./configure
		$ make

	Note: No need to run install target

Running
=======

	Run directly from package root dir after make:

	$ ./sktest -d www.google.com -n 10 -H "Test: value1" -H "Test: value2"

Testing
=======
	Had no time to automate tests. I manually tried to give wrong values to
arguments while running the program under valgrind tool. Additionally I used
Wireshark to make sure custom headers are properly added to the HTTP request.
	If I had more time I would create unit tests for HttpHeader class and
also for median calculation at least.

Design considerations
=====================
	Library public API consists of the following classes:

	HttpConnectivityTest - a single test which can be run separately.

	ConnectivitytestRunner - controls execution of multiple tests with statistics
		calculation functionality. Option for how many parallel tests to run
		should go to this class.

	HttpHeader - a simple wrapper for HTTP header.

	Library usage is demonstrated in the "application" app/main.cpp

	Only headers files under lib/include/public are to be installed (Makefile TODO).

	From the very beginning I had an idea to allow tests to run in parallel, for
example 10 tests in total, 3 in parallel which would give more close to real
world network usage pattern. I would also change the code to open as many Curl
handles as there are parallel tests to save resources.
	Current median calculation mechanism is also sub-optimal because it uses
full sorting O(N*log(N) complexity while proper median algorithms are of linear
complexity.

Future improvements
===================
	Handle HTTP redirects, as with www.google.com example it usually redirects to
a country-local domain (like www.google.com -> www.google.lv in Latvia). It would
give more close to real life usage patterns.
	Add separate IPv4/IPv6 tests. This is a very interesting statistics to see
how IPv4/IPv6 connectivity quality differs for the same resources.
	HTTPS support (HttpConnectivityTest is partially ready, but not tested).
	Disable Curl features which may affect the statistics, like DNS caching.
	Make requests parallel as already said and use Curl async API to avoid
spawning too many threads.