ACM通信

各级政府是数据的主要收集者和提供者。我们的项目专注于通过广域网络(WANs)收集数据，并解决基于internet的大规模数据收集应用程序中出现的可伸缩性问题。此外，由于需要数据的私密性和完整性，安全性是使用公共基础设施(如Internet)的数据收集应用程序的中心问题。许多数字政府应用程序需要在WANs[5]上收集数据。这种应用的一个引人注目的例子是国内税务局电子提交所得税表格。其他数字政府应用包括收集人口普查数据、联邦统计数据和调查;收集和点票电子选票;为美国司法部收集犯罪数据;从传感器收集数据，用于灾害响应应用;从地质调查中收集数据; collecting electronic filings of patents, permits, and securities (for SEC) applications; grant proposals and contract bids submissions; and so on. All these applications have scalability and security needs in common.The poor performance that may be experienced by current digital government users, given the existing state of technology (as in Figure 1a), is largely due to how (independent) data transfers using TCP/IP work over the Internet. TCP/IP is good at equally sharing bandwidth between data streams, which in large-scale applications can lead to poor performance for individual clients (as they receive only a very small share of this bandwidth). Given that TCP/IP is here to stay for the foreseeable future, what is needed is a scalable yet cost- effective solution that can be easily deployed over the existing Internet technology.We are designing and developing a system called Bistro, which addresses the scalability needs of digital government data collection applications while allowing them to share the same infrastructure and resources efficiently, cost-effectively, and securely [1]. Bistro's basic approach is to introduce intermediate hosts---bistros---which allow replacement of a traditionally "synchronized client push" approach with a "nonsynchronized combination of client-push and server-pull" approach (as depicted in Figure 1b). This in turn allows spreading of the workload on the destination server and the network over time, with subsequent elimination of hot spots as well as significant improvements in performance for both clients and servers. Our ongoing research [2, 4] indicates that orders of magnitude of improvement can be achieved with the Bistro architecture and the corresponding data collection algorithms it affords.Bistro's design allows for a gradual deployment and experimentation over the Internet (by simply downloading Bistro server software and installing it on public servers). Bistro's security protocol and trust structure [3] are designed such that only encrypted data travels through (not necessarily trusted) bistros. This means a government agency does not need to trust bistros installed by other agencies or commercial institutions. At the same time, these (untrusted) bistros can significantly improve the agency's data collection performance. Each application (within each agency) can have its own scalability, security, fault tolerance, and other data collection needs, and these applications and agencies can still share available resources, if so desired, across all Bistro servers.We believe an appropriately designed single infrastructure such as Bistro can address all digital government wide-area data collection needs in a scalable, secure, and cost-effective manner. (For more information, see bourbon.usc.edu/iml/bistro/.