Multi-Tenant Multi-Objective Network Slicing Technology In Data Centers

With the continuous development of cloud services at home and abroad,the data center behind the cloud services is also rapidly developing.Resources in data cen-ters such as computing,storage,and networking are all virtualized,which provide cus-tomized services for tenants and ensure their resource requirements.For cloud service companies,customized services can attract more SMEs and users to transfer their IT services to the cloud.For tenants,they don't have to pay attention to the IT environment deployment,reduce maintenance costs,and can concentrate on business development.In this paper,we study the problem about multi-tenant multi-objective bandwidth allocation in data center networks.The objective here refers to the metrics like data flow delay,flow average completion time,etc.By optimizing the network bandwidth resources allocation,we can improve the utilization of data center networks and fur-ther the quality of upper-level IT services.Bandwidth allocation in datacenters should support both performance isolation among tenant and objective-oriented optimization among flows within the same tenant.We use a tenant-objective slice to denote all flows of a tenant that shares the same objective.Each slice consists of multiple tunnels.Each tunnel accommodates all data flows between a pair of virtual machines within a given slice.We propose Stacked Congestion Control(SCC),a shim layer running in the host hypervisor.SCC is a distributed host-based bandwidth allocation design.SCC consists of two layers,where an underlay congestion control(UCC)layer handles contention among slices,and a private congestion control(PCC)layer for each slice to optimize its performance objective.Each slice is assigned a given weight Ws by a certain strat-egy.The upper and under layers are bridged by a tunnel inside the slice,and each tunnel is attached a weight W,.By maintaining the sum of all tunnel weights in the slice equal to the slice weight ?(?)t?s Wt= Ws,UCC achieves weighted bandwidth sharing between slices in a distributed and transparent way.At the same time,UCC provides a unified congestion control strategy for the underlying network and solves the problem of mu-tual interference when coexistence of different objective protocols.PCC dynamically calculates the weight of the tunnel according to the objective characteristics of the data flow in the tunnel,and then implements the dynamic bandwidth allocation among the tunnels in the slice to achieve the objective optimization.By adding a rate-limiting send queue in the ingress of each tunnel,mechanisms between performance isolation among slices and objective scheduling for each slice are completely decoupled,which makes our framework more flexible and extensible.Finally,We evaluate SCC both on a small-scale testbed and with large-scale NS-2 simulations.Compared to the direct coexistence scenarios,SCC can better solve the problem of coexistence and mutual interference of multi-objective protocols,and at the same time implement bandwidth isolation between tenants according to weight distribution.