A Reactive Measurement Framework

Abstract

Abstract. Often when assessing complex network behavior a single measurement is not enough to gain a solid understanding of the root causes of the behavior. In this initial paper we argue for thinking about "measurement" as a process rather than an event. We introduce reactive measurement (REM), which is a technique in which one measurement's results are used to automatically decide what (if any) additional measurements are required to further understand some observed phenomenon. While reactive measurement has been used on occasion in measurement studies, what has been lacking is (i) an examination of its general power, and (ii) a generic framework for facilitating fluid use of this approach. We discuss REM's power and sketch an architecture for a system that provides general REM functionality to network researchers. We argue that by enabling the coupling of disparate measurement tools, REM holds great promise for assisting researchers and operators in determining the root causes of network problems and enabling measurement targeted for specific conditions.

Back in control -- An extensible middle-box on your phone

“…Each of them, however, have been developed and optimized -from scratch -for a specific function and cannot run concurrently with other similar apps. 2 We argue for an extensible platform supporting multiple functions in parallel, each of which can be more easily developed by reusing core traffic manipulation code (i.e., an on-board middlebox), and instantiated on demand.…”

Haystack: A Multi-Purpose Mobile Vantage Point in User Space

“…These measurements-taken from the device's natural perspective-could be proactive in an attempt to better understand the current network context. Alternatively, active measurements could be taken reactively [10] based on some observation-e.g., to diagnose slow transfers or delay spikes. Finally, active measurements could explore "what if" sorts of analysis that could in turn be used to tune the device's operation for better performance.…”

“…For example, a TCP connection can be associated with a specific DNS lookup and both can be coupled with the originating application. Further, we design Haystack to be extensible with new analyses and measurements added over time (e.g., by adding new protocol parsers and by supporting advanced measurement methods such as reactive measurements [10]), and new features to attract and educate users (e.g., ad block, malware detection, privacy leak prevention and network troubleshooting).…”

Providing Grid Schedulers with Passive Network Measurements

“…The data collected by such a technique is at best a close estimate of the performance of the surrounding nodes, but is still merely an estimate. REM [14] is a framework that aims to make sense of different performance indicators to identify unexpected network behavior and react by triggering automated network analysis. Although the motivations are different, both REM and GridMAP regard measurement as an automated process rather than an isolated and manual activity.…”