The reading for 8 byte plaintext data is missing because the
ICMP-Header and the timestamp sent by ping6 are together
at least 16 byte long.
The chart shows a linear increase of round-trip time
with jumps occurring approximately every 100 bytes.
These spikes can be attributed to the 128 byte maximum
link layer frame size defined by IEEE 802.15.4 which includes
header and trailer. These jumps occur earlier when
sending DTLS protected packets due to the additional DTLS
packet headers, the HMAC size and the explicit Initialization
Vector in each packet. See Section 3 for more details
on the packet structure.
Both the increased packet size and processing overhead
lead to an increased end-to-end transmission latency for
DTLS packets compared to plaintext packets. In the single
hop scenario, transmission latency was increased by up
to 95 ms for AES-128 and up to 75 ms for SHA-1 encryption
which were an average increase of 62% and 35% respectively
over the plaintext case. In the multi hop scenario,
round trip times increased by a maximum of 163 ms and
were 74% longer on average for AES-128 encrypted packets.
Packets with a SHA-1 HMAC took up to 129 ms longer
for the round-trip with an average of 40% more time being
spent. The decreased performance for transmission latency
is mostly due to the large packet overhead of up to 64 bytes
which consists of 13 byte DTLS record header, 16 byte
Initialization Vector, 20 byte HMAC, and up to 15 byte padding.
Calculating a SHA-1 hash of a 255 byte plaintext message
only takes 9 ms, encryption with AES-128 takes
another 12 ms. Both operations do not contribute significantly
to the overall transmission latency. This is consistent
with the measurements for 16-byte plaintext (RTT of
58 ms) which increases to 90 ms with AES-128. Including
the overhead of the DTLS record format, 16 plaintext bytes
are expanded to a 77 byte message. Sending 80 bytes via
ping requires 78 ms which indicates a computational
overhead of around 12 ms in this case. A more detailed
analysis of the transmission overhead from an energy perspective
is provided in Section 5.4.
5.2. Handshake latency
Another performance indicator to consider is the latency
introduced by performing a DTLS handshake. We