next up previous
Next: 6 Conclusion Up: Network Aware Data Transmission Previous: 4.4 Analysis

5 Related Work

Network-aware applications are a hot research area which has been widely discussed. This type of application can be classified by their adaptation behavior. [24] gives a discussion about adaptation models used by network aware application, and [1] presented a framework-based approach to develop network aware applications. Some related work in the mobile computing environment is implemented in Odyssey [5,16,20,21].

Compression before transmission and related pre-processing techniques have been used by many network applications [8,9,10,14,26]. Among them, [8] and [14] discussed two examples which are very similar with our work. [8] talks about the trade-off between the compression ration and the compression time. It presents a system to automatically and dynamically select the compression format to reduce the Total Delay based on the future resource performance. Similar to their work, which uses NWS[25] to detect network performance, we also uses an existing network monitoring system to help predict network performance. But we focus on the judgment whether or not to use compression, not compression format, since compression does not necessarily improve the application's performance.

[14] talks about how to use compression techniques in a transcoding proxy in a mobile network environment. By predicting transcoding delay, transcoding size and network bandwidth, it determines whether to transcode and how much to transcode an image for store-and-forward transcoding and streamed transcoding. The problem they focus on is similar to ours, trying to decide which data to send onto the network, but they uses a different way to monitor and predict the network performance, which is very similar with that of [23].

Neither of these works consider the possibility of overlap among different processing modules. Computing the total execution time as the simple arithmetic summary of each module's execution time, in many cases, can impair application performance.

As a key component of the application, available bandwidth prediction is a hot research topic. IDMaps[6] suggests a scalable Internet-wide architecture, which measures and disseminates distance information on the global Internet. NWS[25] is trying to provide accurate forecasts of dynamically changing performance characteristics for a distributed set of metacomputing resources. SPAND[23] determines network characteristics by making shared, passive measurements from a collection of hosts. NIMI[17] proposes to deal with this problem from the perspective of infrastructure, trying to provide a large-scale, extensible platform for network measurement. Besides these systems, which need complicated configuration, simple tools like bprobe/cprobe[2], nettimer[12], pathchar[11] and sting[22] are also available for network performance measurement.


next up previous
Next: 6 Conclusion Up: Network Aware Data Transmission Previous: 4.4 Analysis
root 2001-10-09