Research On Data Plane Update Strategy In Data Center Networks

The data center,also known as the "server farm",originated from the service provider,whose primary goal is to reduce the management cost of servers and ap-plications for users.In recent years,with the large-scale growth of Internet users and the emergence of virtualization technology and commercial servers,some well-known international network services companies such as Google and Microsoft have established their own data center for storing large-scale data.Taking advantage of its distributed storage and powerful processing ability,they can transmit the data to the global user timely and efficiently.Unlike traditional Local Area Network(LAN)and Wide Area Network(WAN),data centers,as resource-intensive infrastructures,need to respond quickly to the requests of a large number of Internet end users and can process multiple computational tasks with massive data in parallel.Software-defined networking technology can meet the new features in data center,and presents tremendous advantages in terms of efficient centralized network management,vir-tual machine automatic deployment and intelligent migration,flexible multi-path forwarding,virtual multi-tenant isolation and security protection.Google and Mi-crosoft use software-defined networking technology to interconnect their data centers and achieve higher network utilization,lower delay,and less packet loss.Software-defined data centers have become an evolution trend of data centers.In this paper,we study the three key issues in software-defined data centers:minimize transient congestion in asynchronous network update,loop-and congestion-free synchronous network update and the design and placement of backup routing in the data plane.The main contributions of this paper are summarized as follows:(1)Firstly,we focus on minimizing transient congestion update problem.In this section we advocate to find an update plan that minimizes the transient con-gestion,given the number of intermediate stages within which the update needs to be done.Our problem is more general in the sense that it allows the operator to navigate a broader design space,where one may trade off update speed,represented by the number of intermediate stages,for the extent of transient congestion.First,We take into account splittable and unsplittable flow as constraints to meet the different application requirements,since applications such as video do not work well when their flows are split.Based on this,we formulate the minimum congestion up-date problem(MCUP)as an optimization program and prove its hardness.Second,we propose two efficient algorithms to solve MCUP.We first propose a randomized rounding algorithm and prove that it yields a O(logk)upper bound of link conges-tion,where k is the number of switches.We further propose a greedy improvement algorithm which has the same approximation ratio O(log k)as the rounding algorith-m in general topologies,and an approximation ratio of 4 for fat-tree in particular.Finally,we evaluate our algorithms in both DCN and WAN scenarios using pro-duction topologies.Simulation results show that our algorithms can reduce average link congestion by 60%,and save 30%control overhead compared to prior work.Experimental results on Mininet show that our solution is 50%faster than prior work.(2)Secondly,we focus on loop-and congestion-free network synchronous up-date problem.Motivated by the advent of tightly synchronized SDNs,we initiate the study of algorithms for consistent network updates in "timed SDNs"—SDNs in which individual node updates can be scheduled at specific times.We introduce a natural and new optimization problem for timed SDNs as we aim to find a network update schedule which minimizes the overall network update time,while ensuring loop-freedom and congestion-freedom at any moment in time.We presents Chronus,which is based on provably congestion-and loop-free update scheduling algorithms,and completely avoids the flow table space headroom required by existing two-phase update approaches.We formulate the Minimum Update Time Problem(MUTP)as an optimization program and prove its hardness.We propose a tree algorithm to check the feasibility and a greedy algorithm to find a update sequence in poly-nomial time.Extensive experiments on Mininet and numerical simulations show that Chronus can substantially reduce transient congestion and save over 60%of the rules compared to state of the art.(3)Finally,we design a failure recovery system using backup routing in the data plane.The statistics data from Microsoft' s data center shows that the probability of having at least one link failure within five minutes is 21.5%.We present Sentinel,a novel failure recovery system for traffic engineering that pre-computes and installs backup tunnels to improve the robustness of software defined wide area networks(WANs).When a link fails,switches locally redirect traffic to backup tunnels and recover immediately in the data plane,thus substantially reducing the transient congestion compared to reactive rescaling.On the other hand Sentinel complete-ly avoids the bandwidth headroom required by existing proactive approaches like FFC,and improves efficiency of operating the expensive WAN.We make several technical contributions in designing Sentinel.We formulate traffic engineering with backup tunnels(TE-BT)as optimization programs.We propose an approximation algorithm to efficiently solve the problem.We further present a concrete design and implementation of the system based on Openflow group tables for backup tunnels.Extensive experiments on Mininet and numerical simulations show that similar to FFC,Sentinel reduces congestion by 45%compared with rescaling,and its algorithm runs much faster than FFC.Sentinel only introduces a small number of additional forwarding rules and can be readily implemented on today's Openflow switches.