Analysis of discarding policies in high-speed networks

Networked applications generate messages that are segmented into smaller, fixed or variable size packets, before they are sent through the network. In high-speed networks, acknowledging individual packets is impractical; so when congestion builds up and packets have to be dropped, entire messages are lost. For a message to be useful, all packets comprising it must arrive successfully at the destination. The problem is therefore which packets to discard so that as many complete messages are delivered, and so that congestion is alleviated or avoided altogether. Selective discarding policies, as a means for congestion avoidance, are studied and compared to nondiscarding policies. The partial message discard policy discards packets of tails of corrupted messages. An improvement to this policy is the early message discard that drops entire messages and not just message tails. A common performance measure of network elements is the effective throughput which measures the utilization of the network links but which ignores the application altogether. We adopt a new performance measure-goodput-which reflects the utilization of the network from the application's point of view and thus better describes network behavior. We develop and analyze a model for systems which employ discarding policies. The analysis shows a remarkable performance improvement when any message-based discarding policy is applied, and that the early message discard policy performs better than the others, especially under high load. We compute the optimal parameter setting for maximum goodput at different input loads, and investigate the performance sensitivity to these parameters     

Optimal smoothing schedules for real-time streams

We consider the problem of smoothing real-time streams (such as video streams), where the goal is to reproduce a variable-bandwidth stream remotely, while minimizing bandwidth cost, space requirement, and playback delay. We focus on lossy schedules, where data may be dropped due to limited bandwidth or space. We present the following results. First, we determine the optimal tradeoff between buffer space, smoothing delay, and link bandwidth for lossy smoothing schedules. Specifically, this means that if two of these parameters are given, we can precisely calculate the value for the third which minimizes data loss while avoiding resource wastage. The tradeoff is accomplished by a simple generic algorithm, that allows one some freedom in choosing which data to discard. This algorithm is very easy to implement both at the server and at the client, and it enjoys the nice property that only the server decides which data to discard, and the client needs only to reconstruct the stream.In a second set of results we study the case where different parts of the data have different importance, modeled by assigning a real weight to each packet in the stream. For this setting we use competitive analysis, i.e., we compare the weight delivered by on-line algorithms to the weight of an optimal off-line schedule using the same resources. We prove that a natural greedy algorithm is 4-competitive. We also prove a lower bound of 1.23 on the competitive ratio of any deterministic on-line algorithm. Finally, we give a few experimental results which seem to indicate that smoothing is very effective in practice, and that the greedy algorithm performs very well in the weighted case.Received: 21 November 2001, Accepted: 6 November 2003, Published online: 6 February 2004Research supported in part by Israel Ministry of Science. An extended abstract of this paper appeared in Proc. 19th ACM Symp. on Principles of Distributed Computing, July 2000.