Efficient and Reliable Communications in Distributed Networks

申请者的长期研究目标是使通信(无线或有线)更高效(在消耗电力等资源方面)、更快、更可靠。四个拟议的短期研究计划中有三个集中在广播这一重要的通信原语上，而第四个短期研究计划旨在提高分布式存储系统的效率和可靠性，在分布式存储系统中，我们将存储视为及时与擦除(存储节点故障)进行通信。*数十亿使用无线电进行通信的设备电源有限(例如，它们使用电池)。这些设备可以受益于在诸如广播之类的通信原语中的能量消耗的节省。*此外，更快和更可靠的通信提高了服务质量，这是一个重要的标准，尤其是在实时应用中。我们的第一个短期目标是分析能量积累在降低广播能耗方面的收益。第二个短期目标是改善实时应用的广播服务质量，例如实时视频流。*分布式有线网络也可以从更快的广播中受益。在计算集群和数据中心中，多对多数据传输(广播的泛化)可能会占用很大比例的应用程序执行时间。例如，在MapReduce框架中，被建模为多对多传输的映射器和减少器之间的数据传输可能占总运行时间的50%以上。因此，非常希望设计有效的算法来缓解数据传输瓶颈，例如在MapReduce中广泛报道的那些。*用于在无线和有线网络中寻址擦除的类似技术和思想已经被用于分析和处理分布式存储系统中的擦除。考虑到当今数据中心的规模，存储、带宽和计算要求甚至略低于现有数据中心的算法和代码可以释放大量资源，例如，节省维护成本和电力使用，并可以间接保护我们的环境。该研究的最后一个短期目标是通过在一类称为局部可修复码的删除码中寻找最优二进制码来降低编码/编码的计算复杂度，这种纠删码因其简单性而具有吸引力。

The applicant's long-term research objective is to make communication (wireless or wired) more efficient (in terms of consuming resources such as power), faster, and more reliable. Three out of four proposed short-term research plans are focused on broadcasting, an important communication primitive, while the fourth short-term research plan aims to improve efficiency and reliability of distributed storage systems, where we view storage as communication in time with erasures (storage node failures).******Billions of devices that use radio for communication have limited power supplies (e.g., they run on batteries). Those devices can benefit from saving in energy consumption in communication primitives such as broadcast. ***Also, faster and more reliable communication improves quality of service, an important criteria particularly in real time applications. Our first short-term objective is to analyze the gain of energy accumulation in reducing energy consumption of broadcast. The second short-term objective is to improve quality of service of broadcast for real time applications such as real time video streaming.******Distributed wired networks can also benefit from faster broadcast. In computing clusters and data centers, many-to-many data transfer (a generalization of broadcast), can take a large percentage of applications execution time. For example, in a MapReduce framework, data transfers between mappers and reducers modeled as a many-to-many transfer, can account for more than 50% of the total running time. It is, therefore, highly desirable to devise effective algorithms to alleviate the data transfer bottleneck such as those widely reported in MapReduce.******Similar techniques and ideas used for addressing erasures in wireless and wired networks have been used to analyze and handle erasures in distributed storage systems. Considering the scale of today's data centers, algorithms and codes with even slightly lower storage, bandwidth and computation requirements than the existing ones can free-up significant amount of resources, resulting in, for example, saving in maintenance cost and electricity usage, and can indirectly protect our environment. The last short-term objective of the proposed research is to reduce computational complexity of coding/encoding by finding optimal binary codes within a class of erasure codes called locally repairable codes, attractive for their simplicity.**