This work explores the impact of decisions made at the physical layer on the 
behavior of delay-constrained communication systems.  The focus is on memoryless 
arrivals and channels with memory.  Performance is assessed in terms of the 
stationary distribution of the transmit buffer at the source.  In particular, the 
complementary cumulative distribution function of the queue occupancy is employed 
to characterize optimal block-length and code-rate as functions of the channel 
parameters.  The system model is crafted to ensure a tractable analysis, while 
remaining general enough to accurately capture many realistic scenarios.  This 
initiative sheds new light on the design of delay-sensitive systems.  It also 
introduces questions on the role and availability of feedback in practical 
implementations.  The theoretical results obtained in this work are supported by a 
numerical study.