The fading channels seen in many wireless systems provide a particularly hostile environment for reliable communication. Current metrics for evaluating the performance limits of fading channels have shortcomings. Ergodic capacity, representing the ultimate error-free communications limit, only applies to systems with infinite coding delay. Practical systems are delay-limited and must use finite-length codes. For delay-limited systems /spl epsi/-capacity and delay-limited capacity are typically used to quantify the communications performance. However, /spl epsi/-capacity is not an estimate of error-free performance while delay-limited capacity tends to be an overly conservative measure. We model practical systems as a single server queue and quantify the communications performance as the average throughput through the queue. Throughput is maximized by optimally selecting the transmission rate and power control strategy. Using this approach we arrive at striking conclusions. First, we show that a throughput very close to ergodic capacity can be achieved with a small coding delay. Second, the optimal transmission rate for some systems can be higher than the ergodic capacity of the channel. Third, we demonstrate the notion that power adaptation does not improve communication performance does not hold for delay-limited systems.