INT Based Network-Aware Task Scheduling for Edge Computing
INT Based Network-Aware Task Scheduling for Edge Computing
复制标题
DOI：
10.1109/ipdpsw52791.2021.00131
复制
发表时间：
2021-06
期刊：
影响因子：
--
通讯作者：
B. Shrestha;Richard Cziva;Engin Arslan
中科院分区：
文献类型：
--
作者：
B. Shrestha;Richard Cziva;Engin Arslan
边缘计算通过处理接近数据源的数据，为延迟敏感的应用程序提供低延迟计算。然而，边缘计算中的任务调度并不能免受性能波动的影响，因为网络流量的动态和不可预测性会对终端设备和边缘服务器之间的数据传输性能产生不利影响。在本文中，我们利用带内网络遥测（INT）来收集有关网络状况的细粒度、时间统计数据，并将网络感知纳入边缘计算的任务调度中。与以数十秒的量级收集端口级或流级统计数据的传统网络监控技术不同，INT 通过捕获数据包级粒度的网络遥测数据提供高度准确的网络可见性，从而提供了精确检测网络拥塞的独特机会。我们使用各种工作负载类型和网络拥塞场景进行的实验分析表明，通过在调度任务时优先考虑网络不拥塞（或轻度拥塞）部分的边缘服务器，通过高精度网络遥测增强边缘计算的任务调度，可以将数据传输时间减少高达 40%，并将总任务执行时间减少高达 30%。
Edge computing promises low-latency computation for delay sensitive applications by processing data close to its source. Task scheduling in edge computing is however not immune to performance fluctuations as dynamic and unpredictable nature of network traffic can adversely affect the data transfer performance between end devices and edge servers. In this paper, we leverage In-band Network Telemetry (INT) to gather fine-grained, temporal statistics about network conditions and incorporate network-awareness into task scheduling for edge computing. Unlike legacy network monitoring techniques that collect port-level or flow-level statistics at the order of tens of seconds, INT offers highly accurate network visibility by capturing network telemetry at packet-level granularity, thereby presenting a unique opportunity to detect network congestion precisely. Our experimental analysis using various workload types and network congestion scenarios reveal that enhancing task scheduling of edge computing with high-precision network telemetry can lead up to 40% reduction in data transfer times and up to 30% reduction in total task execution times by favoring edge servers in uncongested (or mildly congested) sections of network when scheduling tasks.