Asymmetric networks are defined to be those in the forward (or downstream) and reverse (or upstream) link speeds may assume different values. These types of networks are becoming more prevalent due to the growing penetration of hybrid fiber coax (HFC) and asymmetric digital subscriber loop (ADSL) systems, in the last mile or local loop. Consequently, it is important to understand and quantify the effect that asymmetry has on the performance of TCP applications. We study the effect of varying the upstream buffer size on TCP performance. We characterize the maximum achievable throughput as a function of the number of upstream buffers, and also compute the minimum downstream buffer size required to achieve full link throughput.

Time-stamp ordering is one of the consistency preserving algorithms that is used in distributed databases. F. Baccelli (1987) has introduced a queueing model that incorporates the fork-join and resequencing synchronization constraints to analyze the algorithm's performance. The power of interpolation approximation technique is illustrated by obtaining extremely good approximations for this rather complex model. The heavy traffic approximations are obtained by showing that this model has the same diffusion limit as a system of parallel fork-join queues. The light traffic limits are obtained by applying the light traffic theory developed by M.I. Reiman and B. Simon (1989). The heavy traffic limits are computed for general arrival and service distributions, but the light traffic limits are restricted to Markovian systems.>