NeTS-NBD: The Internet Measurements Marketplace

点击翻译按钮获取中文摘要

Internet measurements drive improvements in Internet infrastructure and provide the foundation for the Internet performance research. Thus, the availability and quality of measurements is of fundamental importance to continued progress in these areas. There have been a number of research projects, such as IDMaps, NIMI, King, and Internet Weather Service, that collect general Internet measurements, and then answer specific measurement requests by estimating the requested values from the collected generic data. At the same time, a need often arises for a focused, on-demand measurement of a certain feature. As a typical example, a company selecting a content delivery network for accelerating its Web site may need to compare the performance of several content delivery networks with respect to a specific client population. While some platforms (most notably, Scriptroute service deployed on PlanetLab and, commercially, Keynote Systems) offer measuring hosts to serve this need, they can provide only a limited number of measuring hosts, and hence a limited perspective on the Internet performance. It is clearly not feasible for a single company to provide a representative sample of the entire Internet. Worse, having a limited number of well-known measuring hosts allows systems being measured to "game the system", by optimizing specifically for those hosts. Finally, the closed proprietary system for Internet measurements limits user choice of the types of measurements. For example, in HTTP, a measurement of a page download done with and without pipelining and persistent connections, and with different caching settings at the client, would yield very different results. One cannot rely on the prowess of a single company to timely detect all the trends in HTTP (or other Internet applications), and provide the entire menu of various modes of client operation. Given the scale of the Internet and the unpredictability of the needs for particular measurements, these limitations can only be overcome by harnessing the capacity of the Internet at large itself. The principal investigators (PIs) propose an open Internet-wide system, named DipZoom (for "Deep Internet Performance Zoom"), which is based on peer-to-peer principles and which would enable focused measurements and numerous measuring choices. There are two key ideas behind DipZoom. First, rather than trying to build a global-scale measuring platform, the PIs propose a matchmaking service that merely provides "plumbing" to connect measurement providers and requesters. The measurements themselves are left to outside providers. Second, they will use a market approach as an effective regulator of system behavior without rigid built-in control mechanisms. Together, these two ideas would create a marketplace for Internet measurements: an open ecosystem where anyone can offer measurements from their computers and other computing devices, and anyone can request measurements. The hope is that the combination of an open system with market forces will encourage great and unpredictable innovation and diversity in the offered measurements and measuring devices. Broader Impact: Such a system would significantly enhance our ability to gain insight into Internet characteristics, enable companies to optimize and monitor their Internet sites for focused client populations, and would make "gaming the system" infeasible due to the number of potential measuring hosts.As the first step in realizing this vision, the PIs will concentrate on building the case for DipZoom during the current award period. This involves obtaining experimental data that would demonstrate DipZoom's benefits in gaining insight into the Internet performance, and detailed conceptual design for the system that would lay the groundwork for a future DipZoom implementation.