Commodity distributed systems

创建分布式系统是一项重大的工程实践，设计成本往往高达数百万美元，操作成本则高出数倍。事实上，在北美，仅维持现有计算系统运行每年的总成本就超过400亿美元。尽管如此，计算机系统通常是不安全和不可靠的。这个问题的唯一主要原因是，这些系统的复杂性超出了除最好的人类专家之外的所有人的能力。甚至连联网的家用计算机的安装通常也需要大量的系统知识。但有证据表明，这种复杂性是与生俱来的。随着时间的推移，这个问题将变得更糟，而不是更好，随着我们建立更大的系统并相互连接这些系统。我们研究的目标是解决分布式系统操作的这些难题，这些问题目前需要如此深入的知识才能保持系统的正常运行。*我们计划通过让系统自我管理来做到这一点。此外，我们希望我们的系统纯粹基于商品组件，因为我们认为成本是一个主要问题。为了使系统成为商品，我们必须将自己限制在商品硬件上。尽管今天的计算系统可以使用商品硬件，将它们互连的网络部署和运行成本可能很高。因此，我们正在采用商用无线硬件来构建廉价的多跳无线网络。我们的最终设想是，用户只需将计算机设备插入必要的电源插座，它就会根据需要进行自我配置和自我管理，就像其他主要系统(如电话)自我配置一样。这项研究的意义在于，如果实现，它将使用户能够专注于他们的任务，而不是系统的操作。

The creation of a distributed system is a major engineering exercise, often costing many millions of dollars to design, and several times more to operate.   Indeed, the total cost in North America just to keep existing computing systems functioning exceeds $40 billion annually.  In spite of this expense, computer systems are typically insecure and unreliable.   The single major cause of this problem is that the complexity of these systems exceeds the abilities of all but the very best of human experts.  Even the installation of a networked home computer often requires significant systems knowledge.  The evidence suggests that this complexity is inherent.  Over time this problem will get worse not better, as we build larger systems and inter-connect those systems.  The objective of our research is to address these hard problems of distributed-systems operation that currently require such deep knowledge to keep the system functioning correctly.  We plan to do so by having the systems self-manage.  In addition, we wish our systems to be based purely on commodity components, since we consider cost to be a major issue.  To then make the systems a commodity, we must limit ourselves to commodity hardware.  While computing systems today can use commodity hardware, the networks that interconnect them can be expensive to deploy and operate.  Thus we are adapting commodity wireless hardware to build cheap multi-hop wireless networks.Our end-vision is that a user will simply plug computer equipment into the necessary power outlet and it will self-configure and self-manage as required, in the same manner that other major systems, such as telephones, self-configure.  The significance of this research is that, if realized, it will enable users to focus on their tasks, rather than on the operation of the system.