Research On The Key Technologies Of Function Composition Based On Atomic Capability

Current Internet is suffering architecture ossification. Hence it cannot satisfy stringent demands of various network service provision, heterogeneous network carrying, and communication tunnel creation with DiffServ. Recently, Researchers at home and abroad have studies several architectures such as reconfigurable network, the technique of network routing and switch. Open network architecture like reconfigurable network, SDN and NFV focuses on unfolding network core functions which are re-abstracted and decomposed in a fine-grained way. Several such small functions can be composed into a compound function. Due to its improvement in network adaptability and flexibility, open network architecture becomes one of hot research directions in the domain of network architecture. However, there are still several problems to solve based on the current works.(1) The static structure in the current network node is lacking in adaptability due to illogical function abstraction.(2) There is no approach to compose the network functions in a cooperative way.(3) We should elaborately plan how to place network function in open network.(4) Open network requires network functions to migrate in a flexible way, but current static network domain cannot solve such challenge.This dissertation is supported by the National Basic Research Program of China(2012CB315901)--- ―Research on the Architecture of the Reconfigurable Fundamental Information Communication Network‖. Based on the principle of ―structure implies function‖, we study how to reabstract, compose and migrate network function. We first intruce a dynamic structure —— atomic capability stack, and propose an algorithm to create it. Second, we propose an algorithm to create network-wide service chain so that network functions could work coordinatedly. Third, driven by the creation of atomic capability stack and service chain, we study how to place functions into the network so that the network resource can be fully utilized. Finally, to support function migration over network, we propose a distributed algorithm to decide how to migrate switches in a coarse way. The main contributions are as follows.1. Network functional combination is a promising direction to enhance the Internet adaptability. It decomposes the current layered network into fine-grained building blocks and combines them on demand. But it is not clear about which legacy functions should be decomposed and how to combine them in an optimal way. Directed by the principles of Complex Adaptive System, we propose a novel adaptive architecture, i.e., reconfigurable architecture(RECON). First, differing from existing solutions, RECON decomposes functions of the protocol stack at the layer 3 and 4 into atomic capabilities, fine-grained building blocks, to open the network core functions. Second, in an internal node, RECON could customize various atomic capability stacks on demand via combining atomic capabilities in an optimal way. We formulate atomic capability combination into a nonlinear integer optimization problem, with the proposed algorithm to reach appropriate tradeoff between optimal solution and computation cost. Finally, we implement a proof-of-concept for RECON in the network node. And the results are corroborated by several numerical simulations.2. Isolated network functions(also known as middleboxes) are difficult and costly to manage in an optimal fashion due to their hardware-based implementation and proprietary interfaces. The integration of Software-defined networking(SDN) and network functions virtualization(NFV) is promising to address this challenging issue. However, an efficient framework is required to provide cooperative control of network function instances. To this end, a service chain instantiation framework based on NFV and SDN is proposed in this dissertation. First, the network functions are featured with a new abstraction, called atomic capability, which defines the public features of network functions while the core details are hided. A description-language is utilized to conduct contrivers to develop various instances of atomic capabilities. Second, we propose an implementation of service chain consisting of a sequence of atomic capabilities with order constraints. It is instantiated by optimally selecting different function instances over the network. We formulate this service chain instantiation as an integer linear programing problem, with a simulated annealing solver to approach optimal solution. Third, we implement a proof-of-concept for service chain, namely Matchmaker, atop the SDN controller. The experimental results demonstrate that Matchmaker can manage network functions in an efficient and scalable way.3. Today’s Internet hosts a wide variety of network services, which often must be distributed over the Internet to provide and support network-based services and applications. Understanding and exploring this distribution is essential for maintaining the well-being Internet and its applications, because efficient placement of these network services can dramatically reduce the network latency and improve the user utilization. Network service placement is a big challenge though, due to the explosively increasing network service, the limited network resource and high diversity of request patterns. In this paper, we model the network services placement problem as a Nash bargaining game,and propose a cooperative distributed network service placement algorithmic framework based on bargain theory. With simulations under typical scenarios, we show that our strategy can make a tradeoff between network utilization maximization and users’ experience.4. Dynamic Controller Provisioning Problem(DCPP) is a key problem for scalable SDN. Previously, the solution to this problem focused on adapting the number of controllers and their locations with changing network conditions, but ignored balancing control loads via switch migration, which could reduce the load of the maximum-load controller. In this article, we design a scalable control mechanism via switch migration to maximize control resource utilization, and we define this problem as Switch Migration Problem(SMP). First, due to the various demands on controller CPU, bandwidth and memory for different switch, we provide a resource consuming model for SDN that takes the switches as the resource consumers and controllers as the resource providers. Then SMP is reduced to a network utilization maximization problem which is a typical combinational problem. Second, we introduce a synthesizing distributed algorithm to solve SMP--- Distributed Hopping Algorithm(DHA), to disperse the calculations among the controllers involved. Finally, we implement a proof-of-concept for DHA based on Beacon controller, called DHA-CON, and the results are corroborated by several numerical simulations.