Common use of Network Virtualization in Data Centre Networks Clause in Contracts

Network Virtualization in Data Centre Networks. Deliverable D1.3 [2], produced in WP1, which reports on existing optical and control plane technologies has analysed a variety of solutions for network virtualization considering both industrial products and open-source software. In both cases, network virtualization strongly relies on the SDN paradigm. A conceptually centralized SDN controller configures the underlying physical network devices and/or the servers at the network edges to create multiple instances of virtualized and isolated networks which share a common physical infrastructure. The network virtualization solutions analysed in WP1 mostly operates on L2-L3 networks, without considering the virtualization of optical technologies. Two main categories of network virtualization have been identified: overlay-based network virtualization and direct fabric programming. The former solution delivers customer-defined infrastructures, often including L4-L7 services and based on the required application-layer connectivity. The multi-tenant overlay networks are implemented encapsulating the traffic at the end-point level, using VXLAN, NVGRE or STT technologies. The traffic generated at the network edges is delivered through virtual tunnels created on top of the physical infrastructure, without requiring any configuration of the internal network devices. Overlay- based virtualization is simple and fast to deploy, since it requires only configuration at the network edges, without any impact on the configuration of the physical connectivity. However, this is also its major limitation, since the decoupling between overlay and physical networks means lack of visibility and, consequently, lack of control on the data plane resources. In other terms, the physical connectivity is used by the virtual tunnels as it is, without any coordinated adaptation to optimize the network performance and usage, to meet specific requirements at the virtual infrastructure level or just to react to data plane failures. On the other hand, underlay-based network virtualization solutions with direct configuration of the network fabric dynamically create network paths programming the virtual or physical devices, for example using the OpenFlow protocol [OF], and maintain more control on the data plane configuration. The current research is focusing on possible mechanisms to efficiently coordinate these two virtualization approaches. This coordination has twofold benefits: on one hand it allows to reconfigure the underlying data plane depending on the overlay virtual networks that are currently established over the infrastructure and, on the other hand, it allows to plan new overlay networks (or modify the existing ones) depending on the capacity of the connectivity provisioned through the configuration of virtual paths in the data plane. This mixed approach will be adopted also in COSIGN, with the objective of bringing a powerful level of mutual awareness and cooperation between the underlay- based and overlay-based virtualization. In the COSIGN architecture (Figure 2), as initially defined in Deliverable 1.1 [1], network virtualization based on direct programming of network devices is implemented through infrastructure control functions provided at the SDN controller level. This kind of virtualization, applied at the infrastructure level, will compose and deliver virtual network slices with optical capabilities and able to expose a certain level of network programmability. The SDN controller, in fact, will also provide dedicated functions to operate, manage and control each virtual slice, for example for monitoring or provisioning of virtual paths and connectivity. This infrastructure-level virtualization will be entirely addressed in WP3 and matches the Virtual Data Centre use-case defined in WP1.

Network Virtualization in Data Centre Networks. Deliverable D1.3 [2], produced in WP1, which reports on existing optical and control plane technologies has analysed a variety of solutions for network virtualization considering both industrial products and open-source software. In both cases, network virtualization strongly relies on the SDN paradigm. A conceptually centralized SDN controller configures the underlying physical network devices and/or the servers at the network edges to create multiple instances of virtualized and isolated networks which share a common physical infrastructure. The network virtualization solutions analysed in WP1 mostly operates on L2-L3 networks, without considering the virtualization of optical technologies. Two main categories of network virtualization have been identified: overlay-based network virtualization and direct fabric programming. The former solution delivers customer-defined infrastructures, often including L4-L7 services and based on the required application-layer connectivity. The multi-tenant overlay networks are implemented encapsulating the traffic at the end-point level, using VXLAN, NVGRE or STT technologies. The traffic generated at the network edges is delivered through virtual tunnels created on top of the physical infrastructure, without requiring any configuration of the internal network devices. Overlay- based virtualization is simple and fast to deploy, since it requires only configuration at the network edges, without any impact on the configuration of the physical connectivity. However, this is also its major limitation, since the decoupling between overlay and physical networks means lack of visibility and, consequently, lack of control on the data plane resources. In other terms, the physical connectivity is used by the virtual tunnels as it is, without any coordinated adaptation to optimize the network performance and usage, to meet specific requirements at the virtual infrastructure level or just to react to data plane failures. On the other hand, underlay-based network virtualization solutions with direct configuration of the network fabric dynamically create network paths programming the virtual or physical devices, for example using the OpenFlow protocol [OF], and maintain more control on the data plane configuration. The current research is focusing on possible mechanisms to efficiently coordinate these two virtualization approaches. This coordination has twofold benefits: on one hand it allows to reconfigure the underlying data plane depending on the overlay virtual networks that are currently established over the infrastructure and, on the other hand, it allows to plan new overlay networks (or modify the existing ones) depending on the capacity of the connectivity provisioned through the configuration of virtual paths in the data plane. This mixed approach will be adopted also in COSIGN, with the objective of bringing a powerful level of mutual awareness and cooperation between the underlay- based and overlay-based virtualization. In the COSIGN architecture (Figure 2), as initially defined in Deliverable 1.1 [1], network virtualization based on direct programming of network devices is implemented through infrastructure control functions provided at the SDN controller level. This kind of virtualization, applied at the infrastructure level, will compose and deliver virtual network slices with optical capabilities and able to expose a certain level of network programmability. The SDN controller, in fact, will also provide dedicated functions to operate, manage and control each virtual slice, for example for monitoring or provisioning of virtual paths and connectivity. This infrastructure-level virtualization will be entirely addressed in WP3 and matches the Virtual Data Centre use-case defined in WP1. Figure 2. COSIGN architectural blueprint [1] The overlay-based virtualization functions, with the objective of delivery customized overlay networks matching a high-level description of application-layer connectivity requirements (e.g. in terms of traffic profiles) will be addressed in WP4. In COSIGN, the overlay-based virtualization can be dynamically composed with infrastructure-level virtualization, obtaining overlay networks established on top of highly-programmable virtual slices. This approach allows the bi-directional cooperation between the COSIGN architectural layers and functional entities managing overlay-based (WP4) and infrastructure-level (WP3) network virtualization, with all the benefits discussed above for the mixed virtualization model. An additional innovation of the COSIGN project is the adoption of an optical data plane in the DC network, based on a mixed set of heterogeneous technologies. In order to efficiently manage the whole infrastructure through unified procedures, COSIGN architecture includes a further layer of virtualization mechanisms applied at resource level. The virtualization of optical resources allows for generating an abstracted and uniform view of the different optical technologies available at the data plane, while maintaining a powerful description of resource capabilities. This unified model simplifies the manipulation of the heterogeneous resources and constitutes a fundamental enabler for the infrastructure-level virtualization responsible to create virtual optical slices. Optical resource virtualization is discussed in further details in the following subsection.

Network Virtualization in Data Centre Networks. Deliverable D1.3 [2], produced in WP1, which reports on existing optical and control plane technologies has analysed a variety of solutions for network virtualization considering both industrial products and open-source software. In both cases, network virtualization strongly relies on the SDN paradigm. A conceptually centralized SDN controller configures the underlying physical network devices and/or the servers at the network edges to create multiple instances of virtualized and isolated networks which share a common physical infrastructure. The network virtualization solutions analysed in WP1 mostly operates on L2-L3 networks, without considering the virtualization of optical technologies. Two main categories of network virtualization have been identified: overlay-based network virtualization and direct fabric programming. The former solution delivers customer-defined infrastructures, often including L4-L7 services and based on the required application-layer connectivity. The multi-tenant overlay networks are implemented encapsulating the traffic at the end-point level, using VXLAN, NVGRE or STT technologies. The traffic generated at the network edges is delivered through virtual tunnels created on top of the physical infrastructure, without requiring any configuration of the internal network devices. Overlay- based virtualization is simple and fast to deploy, since it requires only configuration at the network edges, without any impact on the configuration of the physical connectivity. However, this is also its major limitation, since the decoupling between overlay and physical networks means lack of visibility and, consequently, lack of control on the data plane resources. In other terms, the physical connectivity is used by the virtual tunnels as it is, without any coordinated adaptation to optimize the network performance and usage, to meet specific requirements at the virtual infrastructure level or just to react to data plane failures. On the other hand, underlay-based network virtualization solutions with direct configuration of the network fabric dynamically create network paths programming the virtual or physical devices, for example using the OpenFlow protocol [OF], and maintain more control on the data plane configuration. The current research is focusing on possible mechanisms to efficiently coordinate these two virtualization approaches. This coordination has twofold benefits: on one hand it allows to reconfigure the underlying data plane depending on the overlay virtual networks that are currently established over the infrastructure and, on the other hand, it allows to plan new overlay networks (or modify the existing ones) depending on the capacity of the connectivity provisioned through the configuration of virtual paths in the data plane. This mixed approach will be adopted also in COSIGN, with the objective of bringing a powerful level of mutual awareness and cooperation between the underlay- based and overlay-based virtualization. In the COSIGN architecture (Figure 2), as initially defined in Deliverable 1.1 [1], network virtualization based on direct programming of network devices is implemented through infrastructure control functions provided at the SDN controller level. This kind of virtualization, applied at the infrastructure level, will compose and deliver virtual network slices with optical capabilities and able to expose a certain level of network programmability. The SDN controller, in fact, will also provide dedicated functions to operate, manage and control each virtual slice, for example for monitoring or provisioning of virtual paths and connectivity. This infrastructure-level virtualization will be entirely addressed in WP3 and matches the Virtual Data Centre use-use case defined in WP1. Figure 2 – COSIGN architectural blueprint [1] The overlay-based virtualization functions, with the objective of delivery customized overlay networks matching a high-level description of application-layer connectivity requirements (e.g. in terms of traffic profiles) will be addressed in WP4. In COSIGN, the overlay-based virtualization can be dynamically composed with infrastructure-level virtualization, obtaining overlay networks established on top of highly-programmable virtual slices. This approach allows the bi-directional cooperation between the COSIGN architectural layers and functional entities managing overlay-based (WP4) and infrastructure-level (WP3) network virtualization, with all the benefits discussed above for the mixed virtualization model. An additional innovation of the COSIGN project is the adoption of an optical data plane in the DC network, based on a mixed set of heterogeneous technologies. In order to efficiently manage the whole infrastructure through unified procedures, COSIGN architecture includes a further layer of virtualization mechanisms applied at resource level. The virtualization of optical resources allows for generating an abstracted and uniform view of the different optical technologies available at the data plane, while maintaining a powerful description of resource capabilities. This unified model simplifies the manipulation of the heterogeneous resources and constitutes a fundamental enabler for the infrastructure-level virtualization responsible to create virtual optical slices. Optical resource virtualization is discussed in further details in the following subsection.