CNS Core: Medium: Distributed Runtime Dataplane Telemetry as an Adaptive Query Scheduling Problem: Algorithms and Applications

必须对构成互联网的网络进行监控，以确保向最终用户提供高质量、可靠的服务。新兴的网络遥测系统依赖于现代数据平面技术（例如，可编程交换机），提供了对当今网络的高度可见性，并有望实现新一代网络安全和网络性能应用（例如，检测新的网络攻击；支持视频流的体验质量要求）。这些系统考虑定义为查询并编译到交换机上的任务。然后，交换机仅对满足给定查询的那些包以行速率执行包处理。然而，现有的网络遥测系统通常是为静态查询和流量工作负载而设计的，不随查询数量或流量速率而扩展，并且假设任务很简单（例如，无状态、单交换机）。该项目旨在开发一类下一代网络遥测系统来解决这些挑战。该项目利用了运行时可编程数据平面设备的出现，旨在开发和实验评估可扩展的遥测系统，该系统可以适应流量和查询动态，并支持跨网络多个交换机的自适应遥测应用。这些设备允许对有限的交换机资源进行时分多路复用，这反过来又促进了细粒度的资源管理，以应对动态并实现可伸缩性。该项目的新科学贡献包括(i)基于多目标近似的查询调度方案，该方案可以在控制准确性-延迟-报告负载权衡的同时管理多个阶段的单个交换机上的资源；（ii）扩展调度方案，以无缝支持需要跨多个交换机的流量可见性的查询；（iii）支持自适应遥测应用程序，以促进其有状态和迭代的执行和测试。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The networks comprising the Internet must be monitored to ensure high quality, reliable service to end users. Emerging network telemetry systems that rely on modern data plane technologies (e.g., programmable switches) offer a high degree of visibility into todays’ networks and promise to enable a new generation of network security and network performance applications (e.g., detecting new cyberattacks; supporting quality-of-experience requirements for video streaming). These systems consider tasks that are defined as queries and compiled onto a switch. The switch then performs packet processing at line rate for only those packets that satisfy the given query. However, existing network telemetry system are typically designed for static query and traffic workloads, don’t scale with the number of queries or traffic rate, and assume that tasks are simple (e.g., stateless, single switch). This project seeks to develop a class of next-generation network telemetry systems that address these challenges. The emergence of runtime programmable data plane devices is leveraged by this project, which seeks to develop and experimentally evaluate a scalable telemetry system that can accommodate traffic and query dynamics and support adaptive telemetry applications over multiple switches across a network. These devices allow for time-division-multiplexing of limited switch resources which in turn facilitates fine-grained resource management to cope with dynamics and achieve scalability. The novel scientific contributions of this project include (i) a multi-objective approximation-based query scheduling scheme to manage resources on a single switch with multiple stages while controlling accuracy-latency-reporting load tradeoff; (ii) an extension of the scheduling scheme to seamlessly support queries that require traffic visibility across multiple switches; and (iii) support for adaptive telemetry applications to facilitate their stateful and iterative execution, and testing.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.