The Research Of The Throughput-optimal Strategies For Caching And Computing Constrained Wireless Networks

To embrace ubiquitous cloud computing at the edge of Internet,wireless networks have evolved for enabling in-network caching and in-network computing.Nonetheless,the efficiency of existing scheduling mechanisms perhaps are not well-qualified to guarantee the performance of network throughputs,since the networking paradigm manages communications and computations separately.In this dissertation,the above intelligent communication,caching,and computation integrated architectures are studied with the aim of maximizing network throughput,in terms of the scheduling policies in two time-scales as well as the networking protocol used in current networks.First,we re-address the general form of content placement problems in in-network caching policy design,in which popular files are stored in the proximity of users for predictable traffic offloading.Next,we turn to scheduling the unpredictable computation tasks,of which the target is to maximize the number of concurrently users served by edge nodes.At last,the networking protocol,i.e.,Internet Protocol(IP),to connect edge nodes is optimized such that the scheduling overhead can be significantly reduced.The main contributions are summarized as follows:1.The in-network file caching policy is optimized to increase the expected throughput during a scheduling period.We re-address the well-known content placement problem in congestible networks of arbitrary topology as an NP-hard integer programming.The issues of the simplified network topology are identified as the main causes of lowthroughput scheduling policies.We prove the well-known submodularity does not hold in this general case;and propose a Joint Caching and Routing of Arbitrary Topology(Jo CRAT)algorithm with a sublinear approximation ratio based on randomized rounding.The Jo CRAT algorithm achieves nearly two times throughput of the classic schemes in the bandwidth-limited evaluations.2.The real-time computational task scheduling policy is optimized to increase the instaneous network throughput.The most of previous works ignore the impact of network topology on throughput maximization problem.A few important special cases for task schedulings are identified with the optimal structures and the exact algorithms,respectively.As for the general case which is proved to be NP-hard,we further develop a distributed approximation Joint Caching,Processing and Routing of Arbitrary Topology(Jo Ca PRAT)algorithm with a constant approximation ratio based on alternative rounding.The empirical evaluations show that our proposed Jo Ca PRAT is with one order of magnitude lower time complexity while yielding overwhelming throughput performance of the optima compared to the classic methods.3.We optimize the scheduling frameworks to embrace the integration of in-network caching and computation to reduce the extra overheads in the conventional IP-based architecture,e.g.,Domain Name Service.In particular,we develop an scheduling framework to natively support in-network caching and computation based on objectoriented networking protocol(OON)technology instead of an IP-based one.The forwarding plane and the class-based naming mechanisms are co-designed for in-network data processing.A prototype of the proposed OON-based framework is implemented in NS-3 platform.The experimental results reveal that in-network transcoding over the proposed framework indeed reduce the average bandwidth consumption for Wi-Fi networks,which may in turn improve the throughputs.