Designing Virtualization Using Optimized Cloud Networks
AT&T is making some key paradigm shifts with its Domain 2.0 design framework for designing virtualized services that run on an optimized deterministic cloud. This agile and adaptable design framework enables AT&T to rapidly build real-time services such as voice, video and messaging on demand. It pivots from the concept of one universal platform for all real-time services to one where services are instantiated on demand on a common commercial-off-the-shelf (COTS) infrastructure. It pulls together the basic principles of Network Functions Virtualization (NFV) that decouples hardware from software and transforms dedicated network functions into software-based Virtualized Network Functions (VNFs) that can run on a common cloud infrastructure while maintaining service requirements for performance, availability, reliability, and security in a multi-tenant cloud environment. Services are deployed in virtual zones with smaller failure domains, impacting fewer subscribers/customers under failure scenarios.
"Agile methodology and a DevOps model are key factors for rapid design and development of virtualized services. AT&T is working both internally and with vendors on an agile approach"
Many real-time services have special requirements for real-time response times and low latencies. A combination of service level objective constraints defined in service definition templates and orchestration mechanisms enable meeting these performance requirements for virtualized services. The optimized deterministic cloud supports high-performance virtual machines (VMs) that are optimized for CPU and I/O processing as well as optimized networking that supports the guaranteed minimal latency of handling control messages, low latency/jitter, and acceleration technology for supporting guaranteed high throughput for data processing. The optimized deterministic cloud guarantees minimal VM-to-VM latency which fits into end-to-end latency objectives for services. It supports guaranteed network performance and bandwidth allocation to particular VM instances. It supports prioritization of real-time traffic over other traffic by honoring the quality of service constraints specified by the virtual service. Lastly, a key performance requirement of dynamic resource allocation is instantiation and service provisioning in near real-time for the ability to adapt to changing loads or to handle failures.
A key design principle for virtualizing services is decomposition of network functions using Software Defined Networking like concepts into granular VNFs. This enables instantiating and customizing only essential functions as needed for the service, thereby making service delivery more nimble. It provides flexibility of sizing and scaling and also provides flexibility with packaging and deploying VNFs as needed for the service. An optimized deterministic cloud enables deploying latency-sensitive service VNFs close to the users where needed. It enables grouping functions in a common cloud data center to minimize inter-component latency. The VNFs are vendor-agnostic and designed with a goal of being modular and reusable. A virtualized service can be designed using best-in-breed vendors for its granular VNFs.
Decoupling subscriber data and state improves the reliability of virtualized services. It also improves scalability as the data tier can scale independently of the network function. Use of open source technologies like Openstack—a common cloud platform— will play an important role in bringing together industry innovation in the design of virtualized services. Equally important is the use of a common set of technologies such as virtualized load balancers and firewalls that can be instantiated and managed in a common way. Virtualized services are composed of VNFs and common components. They are designed to be agnostic of the location to leverage capacity where it exists in the optimized deterministic cloud. Services are instantiated in any location that meets the performance and latency requirements of the service.
The design of virtualized services requires that VNFs must be only software based and must support virtualization. They must support the ability to execute on an AT&T cloud platform and should be de-coupled from the hardware supported in the AT&T cloud. Open and standard APIs must be supported for exposure to first and third party developers. The set of reusable VNFs forms the basis of a VNF catalog that is made available to service designers via a service design and creation environment. Service-specific custom parameters and QoS policies can be specified for service instantiation within the optimized deterministic cloud.
The Domain 2.0 design framework enables on-demand automated reliable instantiation of virtualized services. The design of virtualized services includes designing policy based auto-recovery from software as well as hardware failures in near real-time. The design takes advantage of cloud elasticity for auto-scaling of service VNFs both on-demand and in near real-time based on performance monitoring. The virtualization design of services specifies adaptive, reactive, and predictive policies for both auto-recovery as well as self-scaling.
Agile methodology and a DevOps model are also key factors for rapid design and development of virtualized services. AT&T is working both internally and with vendors on an agile approach. The granular design and modularity of VNFs play an important role in the ability to create a service iteratively. Use of cloud-based production environments for testing the service design further enables rolling out virtualized services rapidly.
In conclusion, virtualized services are designed to be composed of granular decomposed reusable VNFs and common components. Vendor provided VNFs need to conform to virtualization guidelines. VNFs are designed to be reusable across services. An iterative agile methodology and DevOps model are used for rapid design of services. AT&T’s Domain 2.0 design framework enables achieving the end goal of rapidly designing and delivering services on an optimized deterministic cloud.