Understanding the delay behavior of network coding
with a fixed number of receivers, small field sizes and a limited
number of encoded symbols is a key step towards its applicability
in real-time communication systems with stringent delay
constraints. Previous results are typically asymptotic in nature
and focus mainly on the average delay performance. Seeking to
characterize the complete delay distribution of random linear
network coding, we present a brute-force methodology that is
feasible for up to four receivers, limited field and generation
sizes. The key idea is to fix the pattern of packet erasures and
to try out all possible encodings for various system and channel
parameters. Our findings, which are valid for both decoding delay
and ordered-delivery delay, can be used to optimize network
coding protocols with respect not only to their average but also
to their worst-case performance.