With the growth in Internet access services over networks with asymmetric links such as asymmetric digital subscriber line (ADSL) and cable-based access networks, it becomes crucial to evaluate the performance of TCP/IP over systems in which the bottleneck link speed on the reverse path is considerably slower than that on the forward path. In this paper, we provide guidelines for designing network control mechanisms for supporting TCP/IP. We determine the throughput as a function of buffering, round-trip times, and normalized asymmetry (defined as the ratio of the transmission time of acknowledgment (ACK) in the reverse path to that of data packets in the forward path). We identify three modes of operation which are dependent on the forward buffer size and the normalized asymmetry, and determine the conditions under which the forward link is fully utilized. We also show that drop-from-front discarding of ACKs on the reverse link provides performance advantages over other drop mechanisms in use. Asymmetry increases the TCP already high sensitivity to random packet losses that occur on a time scale faster than the connection round-trip time. We generalize the by-now well-known relation relating the square root of the random loss probability to obtained TCP throughput, originally derived considering only data path congestion. Specifically, random loss leads to significant throughput deterioration when the product of the loss probability, the normalized asymmetry and the square of the bandwidth delay product is large. Congestion in the reverse path adds considerably to TCP unfairness when multiple connections share the reverse bottleneck link. We show how such problems can be alleviated by per-connection buffer and bandwidth allocation on the reverse path.

There has been much interest in using active queue management in routers in order to protect users from connections that are not very responsive to congestion notification. An Internet draft recommends schemes based on random early detection for achieving these goals, to the extent that it is possible, in a system without "per-flow" state. However, a "stateless" system with first-in/first-out (FIFO) queueing is very much handicapped in the degree to which flow isolation and fairness can be achieved. Starting with the observation that a "stateless" system is but one extreme in a spectrum of design choices and that per-flow queueing for a large number of flows is possible, we present active queue management mechanisms that are tailored to provide a high degree of isolation and fairness for TCP connections in a gigabit IP router using per-flow queueing. We show that IP flow state in a router can be bounded if the scheduling discipline used has finite memory, and we investigate the performance implications of different buffer management strategies in such a system. We show that merely using per-flow scheduling is not sufficient to achieve effective isolation and fairness, and it must be combined with appropriate buffer management strategies.

In this paper, we investigate the extent to which fair queueing (and its variants), in conjunction with appropriately tailored buffer management schemes, can be used to achieve the following goals for TCP traffic: (1) alleviate the inherent unfairness of TCP towards connections with long round-trip times, (2) provide isolation when connections using different TCP versions share a bottleneck link (3) provide protection from TCP-unfriendly traffic sources (which might include TCP ACKs since they are not loss-responsive) and misbehaving users, (4) alleviate the effects of ACK compression in the presence of two-way traffic, (5) prevent users experiencing ACK loss (which causes their traffic to be bursty) from significantly affecting other connections, (6) provide low latency to interactive connections which share a bottleneck with "greedy" connections without reducing overall link utilization. The paper proposes new buffer management schemes to be used in conjunction with fair queueing, so as to achieve the above goals for TCP, and compares the performance of the proposed schemes to the performance obtained using random early detection (RED) for packet dropping.

With the envisaged growth in Internet access services over networks with asymmetric links such as asymmetric digital subscriber line (ADSL) and hybrid fiber coax (HFC), it becomes crucial to evaluate the performance of window-based protocols over systems in which the reverse link is considerably slower than the forward link. Even if the actual bandwidth asymmetry is moderate, high effective asymmetries can result because of bidirectional traffic. Our objective is to determine, whether TCP/IP performs reasonably in a setting in which the reverse link is the primary bottleneck. Our main results are as follows. (1) For both the prevalent Tahoe version with Fast Retransmit and the Reno version of TCP we determine the throughput as a function of buffering, round-trip times and normalized asymmetry (taken to be the ratio of the transmission time of ACKs in the reverse path to that of data packets in the forward path). We identify three modes of operation which are dependent on the forward buffer sizes and the normalized asymmetry. (2) Asymmetry increases the TCP's already high sensitivity to random packet losses that might be caused by transient bursts in real-time traffic. Specifically, random loss leads to significant throughput deterioration when the product of the loss probability, the asymmetry and the square of the bandwidth delay product is large. (3) Congestion in the reverse path adds considerably to the TCP's unfairness when multiple connections share the reverse link. Link bandwidth sharing is unfair even for connections with identical round-trip times and hence use of per connection buffer allocation on the reverse path appears essential.