Comment. Monitoring Networked Applications With Incrementak Quantile Estimation.
暂无分享,去创建一个
First of all, we would like to thank the authors for their timely paper focusing on the important area of streaming data. Recently, much attention has been paid by the statistics community to the high dimensionality or massiveness of data in the information technology age. However, streaming data represent the other important feature of the IT age, the high rate of data. Both high dimensionality and high rate require fast computation, but the real-time constraint on streaming data forces its computation to be a magnitude faster than that of the off-line or batch mode of massive data. As a result, in the absence of supercomputers, the algorithms for streaming data have to be very simple to be effective. Chambers et al. deal with streaming data for computer system monitoring. Streaming data arise also in many other fields of science and engineering, such as astronomy, geoscience and sensor networks. Chambers et al. devise a simple and practical algorithm for updating quantiles to be used to monitor the reliability of a large system based on streamed data. Stationarity is implicitly assumed since one could argue that a good computer system should be more or less stable over time until the system is updated. A desirable add-on to the estimated quantile of Chambers et al. is a measure of uncertainty which in the i.i.d. case is trivial because of the relationship between the variance and mean of a binomial random variable. However, it is hard to imagine that a computer system follows an i.i.d. process. The realtime constraint could make the pursuit of an uncer-
[1] Kannan Ramchandran,et al. Microarray image compression: SLOCO and the effect of information loss , 2003, Signal Processing.
[2] Silvia Nittel,et al. Semi-Streaming Quantization for Remote Sensing Data , 2003 .