Optical Network Service Provisioning For Edge Computing

As Internet applications' demand on bandwidth became higher and higher,optical fiber communication networks are playing an important role in cloud computing,which is based on large data centers.At the user side,the optical access network provides end users the broadband access to the Internet;at service side,the optical transport networks take care of the long-distance transmission and the interconnection among large data centers(cloud data centers).The cloud-based converged service model may cause a large amount of propagation delay due to long-distance network transmission,and thus does not apply to service-delay-sensitive services.Driven by ultra-low latency applications such as self-driving cars,the concept of edge computing is attracting more and more attention from the industry.By placing small data centers(edge data centers)at the edge of networks(such as in access networks),edge computing provides users with service resources in close proximity,thereby reducing service latency.Edge computing has changed the convergent service model of cloud computing and has brought new challenges to the optical networks in cloud computing.On the one hand,with edge data centers,delay-sensitive services will no longer need to get services through the transport networks,presenting the trend of "service marginalization" and thus handing over the latency problem to the access networks.On the other hand,edge data centers are widely distributed but their individual capacity is limited.If necessary,they still need to cooperate with each other.This has brought more kinds of communication demands to the optical transport networks that are connected to them,leading to a trend of "interconnection-intensive"at the datacenter side.In order to meet these challenges,this paper focuses on the problem of how optical communication networks support low-latency access and dense interconnection in the context of edge computing.Following the "two trends",this paper proposed four ideas regarding optical network service provisioning for edge computing.First,driven by edge computing,the service model of optical networks will undergo major changes,and the most basic problem is how to design an optical network system structure that can adapt to these changes.For the issue of how edge computing integrates with existing cloud computing and optical network systems,this paper 1)designs an optical transport device model based on edge computing and network function virtualization;2)proposes a hierarchical abstraction and control framework that covers edge/cloud data centers and optical networks.The proposed device model and the control architecture setup the context for studying the problem of provisioning optical network services for edge computing.Second,under the trend of "service marginalization",this paper developes a mixed-integer linear programming model(MILP)for planning virtual machine placement and workload assignment in an edge data center for guaranteeing the application layer latency in edge data center.Base on the above model,this paper further proposes a joint scheding algorithm of TDM-PON upstream bandwidth and tasks in edge data center to further reduce end-to-end service latency for edge computing.Simulation results show that the proposed model and algorithm can effectively reduce the end-to-end delay of about 5%.Third,one manifestation of the "interconnect-intensive" trend is that the collaborative requirements of distributed data centers will lead to more and more virtual optical network services for multi-point interconnection.In order to meet the heterougenous requirements for multi-tenancy virtual optical networks for edge computing,this paper proposes a spectrum optical continuity network-aware virtual optical network mapping scheme for different requirments on spectrum continuity within virtual optical networks.To provide different protections to virtual optical networks,a multi-layer protection scheme was proposed.Results show that the proposed solution can provide network operators with a more accurate virtual optical network service model,which in turn significantly improves the capacity of accommodating virtual optical network.Fourth,another manifestation of the "interconnect-intensive" trend is that the collaborative relationship between data centers will bring more non-immediate communication demands for service migration/backup,ie,advance reservation services.In order to efficiently schedule high-density advance reservation services,this paper designs a cross-stratum reservation scheme for data centers and optical network resources for the initial scheduling of advance reservation services.Targeting on the initial-delay period of scheduled services,this paper also proposed a re-provision algorithm in three dimensions,i.e.,time,spectrum and routing.Simulation and experimental results show that the two-stage solution can accommodate about 3.5%of the advance reservation services and reduce the initial delay.