ITR/SI: Cooperative Computing for Distributed Embedded Systems

本文的目标是为嵌入式系统的大规模网络开发一种集成的计算-通信体系结构，称为协同计算。与分布式计算不同，协同计算是在具有易变节点和连接的动态ad-hoc网络上进行的，并期望在统计上正确执行。协作计算的基本思想是节点以移动的，形成一个全局任务，利用它们的本地资源和状态信息来完成任务的计算或通信。缓存、路由、拓扑发现、对于任何特定的用户定义的应用，可以在协作计算模型上实现为全局任务。 为了有效地协作，所涉及的节点需要知道它们在成功地完成目标任务时所面临的能力和约束。在这样的功率和能力受限的环境中，协作节点必须智能地做出最佳决策以确保结果的质量（QoR）。考虑到这些目标，本文提出的目标是开发一个支持嵌入式系统动态网络协同计算的系统架构，主要研究以下几个问题：如何在应用程序和操作系统级别定义全局任务？如何编程一个非常大的嵌入式系统网络来执行一个全球性的任务，而不单独处理每个节点？如何仅使用对等通信实现自外出？如何衡量一个部分成功的执行？如何预测网络中单个节点和协作节点的功耗和能耗？如何优化一个任务，以减少其在单个节点上和跨网络节点运行的功率和能量戈伊？ 协作计算系统的体系结构是基于智能消息（Smart Messages，SM）的，它可以看作是网络中数据的智能实例。智能消息是通过网络迁移的代码和移动的数据的集合，每次迁移一个网络跳，在每一步执行。智能消息是自出路的，即它们负责确定自己通过网络的路径，利用网络中节点提供的最小设施集，由于节点的资源限制，特别是其邮件戈伊级别的限制，SM可能被拒绝请求发送节点。我们认为，智能消息为各种各样的应用提供了灵活的支持，从数据收集和分发，基于内容的郊游和对象跟踪，涉及其中任务的执行在设备集合上进行的更加分布式的计算应用。 作为这一计划的一部分，我们计划实现和广泛评估智能消息架构的几个原型实现。我们的测试平台将包括基于蓝牙短距离无线通信的嵌入式多控制器和PDA的网络。

The objective of the proposed work is to develop an integrated computational-communication a chitecturefor large scale netwo ks of embedded systems,called cooperative computin .Unlike distributed comput-ing,coope ative computing is pe formed ove a dynamic ad-hoc network with volatile nodes and connections,and expects statistically correct executions.E .ective esource management,in pa ticula power and energymanagement,will be c ucial since most devices in the netwo k will be mobile and will ely on batte y.Thebasic idea in coope ative computing is that nodes coope ate in pe forming a global task in an ad-hoc manneusing thei local esou ces and state information to att act task 's computation o communication.Caching,routing,topology discove y,o any speci .c use -de .ned application,can be implemented as a global taskover the cooperative computing model. To effectively cooperate,the nodes that are involved need to know the esources and constraints they facein successfully accomplishing the ta get task.In such a power and esource constrained environment,thecoope ating nodes must intelligently make the best decisions to ensu e the quality of esult (QoR).With these goals in mind,the p oposed esea ch is to develop a system a chitectu e to support coope ativecomputing ove dynamic netwo ks of embedded systems.Our resea ch will address the following issues:How to de .ne a global task at the application and ope ating system levels?How to prog am very large networks of embedded systems to execute a global task without handlingeach node individually?How to implement self-outing using peer-to-peer communication only?How to measure a pa tially successful execution?How to predict the powe dissipation and energy consumption of individual nodes,and nodes cooper-ating ac oss the network?How to optimize a task to reduce its powe and ene gy budget running on individual nodes,and acrossnodes in network? The system a chitectu e for coope ative computing is based on Smart Messages (SM),which can beviewed as intelligent ca ie s of data in a netwo k.Sma t Messages a e collections of code and mobile datathat migrate th ough the netwo k,a single network hop at a time,executing at each step.Smart Messagesa e self-outing,namely they a e esponsible fo dete mining thei own paths th ough the netwo k,utilizing aminimal set of facilities p ovided by nodes in the netwo k.An SM may be denied a equest to oute th ougha node due to the node 's esou ce limitations,in pa ticula its emaining ene gy level.We conjectu e thatSmart Messages p ovide a .exible support for a wide variety of applications,anging from data collectionand dissemination,content-based outing and object t acking,to more t aditional dist ibuted computingapplications in which execution of a task is sp ead across a collection of devices. As pa t of this p oposal we plan to implement and extensively evaluate seve al p ototype implementationsof the Smart Message a chitectu e.Ou testbed will consist of networks of embedded multicontrollers andPDAs based on Bluetooth sho t-range wi eless communication.