A Co-Located Interference-Aware Static Placement Algorithm In Virtual Network Functions Consolidation

Network function virtualization(NFV)emerging in recent years,by transfering conventional physial hardwares into virtualied network function,running in VMs or containers on x86 commodity servers.NFV technology successfully avoid the specificity and inconvenient installation of traditional physical hardware,provides high agility and flexibility,improves resource utilization while reducing capital expenditures(CAPEX)and operating costs(OPEX),demonstrating great potential for network service providers(ISP).In NFV environment,the request from cloud tenant is abstracted into a service function chain(SFC)which is a sequence of VNF with specific dependencies.The first phase before service running is called VNF static placement phase.ISPs install VNF on server and becomes an operational VNF instance in VM or container,this process is called VNF instantiation.Generally,in order to support thousands of VNF instances required by SFCs in hundreds of servers of a data center,ISPs have to consolidate multiple VNF instances on same server.For cloud service like CISCO ESC,ISPs need to purchase a license for each VNF instantiation.The total costs of VNF instantiation in static placement phase is called instance costs.Under the satisfaction of Quality-of-Service(Qo S)requirements from cloud tenants,ISPs tend to arrange SFCs which require the same type of VNF to share VNF instance for significant instance costs saving.However,recent studies show that VNF consolidation may result in severe performance decline due to physical resource competition,which is called VNF co-located interference(VNF-CI).The placement results of the VNF static placement phase will not only affect the performance of the instantiated VNFs,but will also have a profound impact on the run-time phase when service running.Unfortunately,we currently face two challenges.First,previous works about VNF-CI have mainly focused on run-time phase,the approaches all based on real time data to support prediction and calculation.However,there is no real time data in static placement phase,there is currently no VNF-CI calculation model in static placement phase.Static placement that without VNF-CI calculation model results in randomized initial placement,ISPs need to boot new instance when a instantiated VNF is unable to satisfy Qo S requirement,leading to extra instance costs.The difference in the license price of instantiation of different VNFs can reach hundreds of times at most,and the profits of ISPs may be severely damaged due to being forced to launch expensive new VNF instances.Second,VNF static placement algorithms are all global optimal algorithms.VNF-CI performance decline ratio is related to the packet processing characteristics of each co-located VNF,the calculation complexity of VNF-CI performance decline ratio is proved to be NP-Hard.Existing global optimal algorithms take VNF-CI into consideration will lead to unacceptable computational complexity.We need both calculation model in VNF static placement phase and an effective placement algorithm.To address above issues,in this paper,we first propose a mathematical model for the calculation of accurate VNF-CI decline ratio in static placement phase.Based on the mathematical model,a heuristic static placement algorithm,named as VNF-CIAP,is designed to iteratively place SFCs with minimal instance costs under the satisfaction of Qo S.Experiments show that for the VNF-CI performance decline ratio in static placement phase,the deviation of model estimations and experimental results ranged from 2.30%to 3.17%,which is below the negligible limit.Through extensive simulations,VNF-CIAP saves 9.83% to 14.61% instance costs compared with the ILP-based algorithm.