In this chapter, we introduced the concept of flow-adaptive wireless links that provides service differentiation by adapting link layer error control. The key advantages of our solution is its independence from transport (or higher) layer protocol semantics, and the possibility of co-existence with IPsec. Through measurements in GSM-CSD, we have validated the concept, and show that for fully- reliable flows our solution eliminates all known inefficient cross-layer interactions except for the problem of competing error recovery. In a first study of this prob-lem, we find that even long transient delays on the end-to- end path, do not trigger spurious timeouts in TCP. This made us suspicious that something was wrong with the retransmission timer implemented in TCP-Lite. In Chapter 5, we therefore study in detail both the problem of spurious retransmissions in TCP and TCP-Lite’s retransmission timer. We showed that pure end-to-end error recovery, i.e., when the wireless link is not protected by link layer error recovery, fails as a general solution for optimizing end-to-end throughput when wireless links form parts of the path. Fundamental problems are that the path’s MTU is often too large to yield efficient error recovery, and that network end-points are not capable of dynamically adapting their MTU to changing local error characteristics on (possibly multiple) wireless links. In many cases, this leads to decreased end- to-end throughput, an unfair load on a best-effort network, such as the Internet, and a waste of valuable radio resources. In fact, we show that highly persistent link layer error recovery over wireless links is essential for fully- reliable flows to avoid these problems. Our results also suggest that the GSM-CSD channel is over-protected with FEC, and that the default error recovery persistency standardized for RLP is too low. We show that the through-put of the GSM-CSD channel can be improved by up to 25 percent by increasing the (fixed) RLP frame size to reduce the relative per packet overhead. These results highlight the impor-tance of measurement-based analysis of protocol performance over wireless links.