On the Predictability of Large Transfer TcP ThroughputReport as inadecuate


On the Predictability of Large Transfer TcP Throughput


On the Predictability of Large Transfer TcP Throughput - Download this document for free, or read online. Document in PDF available to download.

With the advent of overlay and peer-to-peer networks,Grid computing, and CDNs, network performance predictionbecomes an essential task. Predicting the throughput of largeTCP transfers, in particular, has attracted much attention. In thiswork, we focus on the design, empirical evaluation, and analysisof TCP throughput predictors for a broad class of applications.We first classify TCP throughput prediction techniques into twocategories: Formula-Based FB and History-Based HB. Withineach class, we develop representative prediction algorithms,which we then evaluate empirically over the RON testbed. FBprediction relies on mathematical models that express the TCPthroughput as a function of the characteristics of the networkpath e.g., RTT, loss rate, available bandwidth. FB predictiondoes not rely on previous TCP transfers in the given path, and itcan be performed with non-intrusive network measurements. Weshow, however, that the FB method is accurate only if the TCPtransfer is window-limited to the point that it does not saturatethe underlying path, and explain the main causes of the predictionerrors. HB techniques predict the throughput of TCP flows froma time series of previous TCP throughput measurements on thesame path, when such a history is available. We show that evensimple HB predictors, such as Moving Average and Holt-Winters,using a history of limited and sporadic samples, can be quiteaccurate. On the negative side, HB predictors are highly pathdependent.Using simple queueing models, we explain the causeof such path dependencies based on two key factors: the load onthe path, and the degree of statistical multiplexing.



College of Computing Technical Reports -



Author: He, Qi - Dovrolis, Constantinos - Ammar, Mostafa H. Mostafa Hamed - -

Source: https://smartech.gatech.edu/







Related documents