The principle of functioning of Net-Entropy relies on the randomness property
of cryptographic algorithms.
A perfectly random infinite Byte string have
a statistical entropy that tends to 8 bits per Byte.
The figure 1 shows the average estimated statistical entropy computed from
small size random messages.

Figure 1: Average statistical entropy estimated from small random messages

The perfect case shown in the figure 1 is not quitely exact in real
world cryptographic applications. This is essentially due to
cryptographic protocols, which insert plain text messages for
connection setup and key establishment. These messages insert a bias in the
Byte distribution of the whole exchanged data, so which decrease the entropy.
The figure 2 shows the entropy of a HTTPS connection (HTTP secured with SSL/TLS).
The key establishment explains the initial low entropy, and the slower growth,
once the content data are ciphered.

Figure 2: Statistical entropy for a HTTPS connection

The figure 3 shows the entropy of a HTTPS connection attacked with an
exploitation of an OpenSSL flaw (BugTraq ID 5363). Data generated by
the attack reduces the connection entropy.

Figure 3: Statistical entropy for an Apache/SSL attack connection

For additional information, the figure 4 shows the entropy of a connection of
plain text protocols such as HTTP, SMTP and TELNET.

Figure 4: Statistical entropy plain data (http, smtp, telnet)