Media Processing in NFV Architectures

Transcription

1 White Paper December 2014 Media Processing in NFV Architectures Introduction Across the IT industry, providers are rapidly moving towards the benefits of agility, flexibility, and cost savings offered by virtualized environments. While data centers and network architects have been investigating Software Defined Networks (SDN) to simplify their networks and reduce operational costs, service and solution providers are evolving their networks to a Network Functions ization (NFV) architecture to realize the network flexibility and infrastructure reuse gained by deploying virtualized workloads and applications across a generic pool of underlying infrastructure, along with the service orchestration and management required to support rapid service deployment. Obviously, costs associated with dedicated infrastructure-per-application can be reduced by moving to fully virtualized infrastructure and applications. At the same time, issues such as cost-benefit tradeoffs, service level agreements (SLAs), quality, and reliability must be addressed before the benefits of a virtualized environment can be fully realized. Real-time applications place special requirements for virtualized environments, and real-time multimedia communications, such as those performed by a Media Resource Function (MRF), place particularly stringent demands on an underlying virtualized infrastructure. This paper briefly reviews the general benefits that NFV architectures offer to service and solution providers and then describes the challenges posed by fully virtualized environments for real-time applications in general, and media processing applications in particular. The paper then shows solutions the industry is developing to meet these challenges and shows how the Radisys virtualized Media Resource Function (vmrf) is ready to deploy as a ized Network Function (VNF) in an NFV environment leveraging VMware s vsphere technology. CONTENTS Why ize? pg. 2 The Drive Toward Network Functions ization pg. 2 Real-time Applications and Media Processing Challenges in an NFV Architecture pg. 2 Radisys vmrf on NFV: a Ready-to-deploy VNF pg. 4 VMware ization Technology for NFV pg. 6 Management and Orchestration of the Radisys vmrf pg. 6 Implications for Cloud Providers pg. 7 Radisys ized Media Processing Solution pg. 8 References pg. 8 Prepared in collaboration with:

2 2 Why ize? The benefits of virtualization are not new to the server and data center industry; however, greater use and an increasing number of deployments of virtualized applications are enabling greater efficiencies and continuing to decrease the total cost of ownership across many industries, including the telecommunications space. In the era of rapid service creation and lifecycle management, creation and deployment of virtualized applications, decoupled from the multiplicity of underlying hardware specifics and host OS variants, is an essential requirement. ization technology enhancements and optimizations from a real-time performance and rapid deployment perspective are enabling new classes of communications services to be cloud-deployable. Complete end-to-end communications infrastructure, as well as end-user applications, allow services to be hosted on general-purpose compute servers. Services can be deployed with a small initial footprint then expanded in near real time, scaled up to meet peak traffic demands and scaled down as traffic slows. This improves the overall flexibility of the underlying compute and network infrastructure to be redeployable for other applications and services. The Drive toward Network Functions ization The NFV architecture was developed by a consortium of service providers to address the special needs and challenges of service providers. In NFV, network node functions are virtualized into building blocks that may be connected into service solutions. By virtualizing network functions, providers hope to create savings in energy, capital investment and operational costs by improving management, resource allocation, scalability, and resilience while at the same time increasing speed of deployment, service agility, and network flexibility. The original white paper, created by an industry specifications group (ISG) within the European Telecommunications Standards Institute (ETSI) in October 2012, set out the benefits, enablers and challenges for NFV as distinct from cloud and SDN. The primary driver and focus of NFV, versus more generic cloud compute architectures and deployments that have been available in the IT industry for well over a decade, is based on carrier deployments in particular the ability to virtualize essentially all aspects of the carrier network. Standardization bodies, particularly ETSI, continue to define the NFV architecture, and NFV now has its own ISG (Industry Specification Group) within ETSI. Dedicated working groups within the ETSI s NFV domain are supported by a wide range of industry participants and contributors, including the following focused work groups: Network Function ization NFV INF (Architecture of the ization Infrastructure) NFV MAN (Management and Orchestration) NFV NOC (Network Operators Council) NFV PER (Performance and Portability) NFV REL (Reliability and Availability) NFV SEC (Security) NFV SWA (Software Architecture) NFV TSC (Technical Steering Committee) Real-Time Applications and Media Processing Challenges in an NFV Architecture In the ongoing definition of NFV, particular focus is being placed on ensuring that real-time communication applications are adequately accommodated within the architecture. The capacity and responsiveness of the application must not be overly degraded by running on top of a virtualized infrastructure that includes hypervisors, host operating systems, and guest operating systems vis à vis applications running on dedicated bare metal deployments. A Media Resource Function (MRF), as defined in IMS and 3GPP Architectures, provides key media processing capabilities for a diverse set of communications applications in telecom as well as enterprise unified communications systems. Example services include multimedia multiparty conferencing, telcom network services (such as network tones and announcements), transcoding for VoLTE, VoWiFi, or WebRTC services, interactive voice and video response (IVVR), multimedia recording or streaming in contact centers or in Lawful Intercept use cases. As for other real-time applications, it is important for an MRF to have uncompromised access to underlying network and processing resources. But for real-time media processing applications, it is even more important to ensure that the quality and reliability of the media is not degraded as compared

3 3 to non-virtualized, bare-metal deployments or as compared to Media Resource Functions (MRFs) deployed on dedicated, purpose-built hardware. Real-time IP media processing requires hard real-time response with low delay and jitter performance. Some of the key quality and reliability metrics expected of a real-time communication network element such as an MRF are: Predictable media processing latency and response times. A standard metric is 5 msec packetization for all active call sessions being processed. For example, if an MRF instance is concurrently processing 4000 active media-enabled callers, then each of the 4000 sessions must be concurrently processed with no more than 5 msec packetization latency. Media jitter management that adapts in real time to varying network conditions. This feature is an essential media processing capability for IP networks because the MRF supports endpoints and callers from a diverse set of access networks; for example, dedicated to unmanaged wired IP networks and mobile networks from managed LTE to unmanaged WiFi networks. Adaptive performance in real time to varying workloads and varying media processing use cases. A hard realtime application such as the MRF must handle varying call scenarios for instance from low bandwidth to high bandwidth with high compute utilization varying drastically over short periods of time. As a result, dynamic compute core-allocation, memory allocation schemes, and network I/O need rapid and adaptive reallocation to provide the greatest capacity of the given server without sacrificing the media quality expected from the MRF. Reliability and redundancy. The MRF may be expected to meet the carrier-grade availability requirements typically sought for real-time communications services, as opposed to best-effort Over The Top (OTT) services. Unlike many other classes of applications, services like IP media processing, which involve a certain kind of real-time workload, cannot be commercialized unless the requisite level of quality of service is in place. This has implications for the underlying virtualization technology, because quality of service characteristics such as latency and jitter must be considered in the aggregate, horizontally across the entire solution stack. Long Term Evolution (LTE) mobile applications and services have a well-defined Quality of Service (QoS) mandated by ETSI through the QoS Classification Indicator (QCI). The QCI describes the response time required for various types of classes of services offered over the LTE network, categorized by the real-time or near real-time nature of these services. Services are classified as being Guaranteed Bit Rate (GBR) or non Guaranteed Bit Rate (non-gbr). Table 1 shows the standardized QCI characteristics. Table 1: Courtesy of ETSI TS Release 12 (Table 6.1.7)

4 4 At the same time, as IP media processing moves further away from bare-metal hardware, challenges in providing hard realtime performance increase, particularly in fully virtualized environments. Adaptive processing, for example, requires a highly optimized virtual machine (VM) and hypervisor. The dynamic compute core-allocation, memory allocation schemes, and network I/O required by the MRF must be assisted by a highly tuned VM/hypervisor with associated I/O optimization for example, NIC partitioning and Single Root I/O ization (SR-IOV). Similarly, workload sharing, which is one of the premises of infrastructure reusability, must protect the real-time processing response. In a virtualized environment, the virtualized MRF running on top of a given compute server may be running alongside a completely different application with compute, I/O, and storage needs of its own. Even so, the MRF, which is supporting real-time communications applications with high sensitivity to real-time processing response, must not be affected by other non or soft real-time applications running alongside the MRF in another virtual machine on the same physical server. Elasticity of resources is a fundamental requirement and driver for virtualized resources, particularly for cloud deployments. Implications for real-time applications become evident as elasticity of resource in the real-time media processing domain pose challenges. This is where the underlying virtualized infrastructure needs to provide elasticity and scalability meeting the hard real-time requirements of media processing elements such as the MRF. The Radisys vmrf on NFV: a Ready-to-deploy VNF The NFV architecture separates the Network Function ization Infrastructure (NFVI) and Network Functions (VNFs). The NFVI includes the virtualization layer (for example, the layer supplied by VMware), to support multiple VNFs running over a common NFVI. The NFVI and VNF components are shown in the simplified NFV architecture diagram in Figure 1 below. An IMS has many functions defined in the 3GPP standards all of which are deployable as virtualized functions in NFV. In Figure 1, each IMS VNF is shown in a different color, representing the fact that each IMS element may be provided by a different vendor. For the purposes of this discussion, the 4 key elements related to IMS service delivery are as follows: Home Subscriber System (HSS). The HSS is a database that maintains subscriber-related profiles and all of the information required to permit session establishment, such as subscriber location and IP address. The HSS also performs authentication and authorization tasks for the IMS entities processing the calls. Call State Control Function (CSCF). The CSCF is usually a SIP server providing a point of contact in the signaling plane, and may also provide session control and basic call processing involving media processing requests to the ized Network Functions (VNFs) EMS vhss EMS EMS EMS vcscf vtas vmrf Computing Storage Network ization Layer Hardware Resources Computing Storage Network Network Functions ization Infrastructure (NFVI) Figure 1: ized IMS in simplified NFV architecture.

5 5 MRF for playing prompts and collecting digits. The CSCF communicates with the HSS using the DIAMETER protocol, which is an evolution of RADIUS. Telecom Application Server (TAS) The TAS provides supplementary voice and video services for SIP subscribers. While CSCF may perform basic session control, enhanced services, such as Voice Call Continuity (VCC), conferencing, or video adverts, are usually performed by a TAS. Media Resource Function (MRF) The MRF provides the shared IP media processing resource for IP audio and video packet streams. The MRF is controlled by the CSCF and TAS servers. Of the VNFs present in a virtualized IMS, the vmrf arguably has the most stringent real-time requirements. It is the only element processing bearer media streams, as opposed to the call control and SIP signaling functions of the CSCF and TAS and the database functions of the HSS. In Figure 1, a red dotted line defines the boundary of the virtualized MRF function and the underlying NFVI. One of the key benefits of virtualization is the ability to quickly turn up capacity as required and then turn capacity off when not required. To achieve this, each virtualized function could be packaged in Open ization Format (OVF) an open, secure, portable, efficient and extensible format for packaging and distributing software to be run in virtual machines. Figure 2 shows how two vmrf VNFs could be deployed as two OVF packages. The red dotted line maps the functionality boundary in the NFV architecture view in Figure 1, to the implementation details shown in Figure 2. The intent is that each vendor would supply their IMS functionality for VNF deployment as an OVF package containing the application software, host operating system, as well as configuration defining its requirements for virtualized hardware, networking, and storage resources. During runtime, the management software monitors demand for various IMS services, and then turns up or turns down OVF packages of virtualized MRFs as required. These OVF packages would run on the VMware virtualization layer, which abstracts the physical computing, network, and storage resources on the Intel x86 hardware, maintaining the virtualized computing, storage, and network resources to run a vmrf instance, in parallel with all the other VNFs running on the virtualization layer at the same time. Historically, to deliver high performance IP media processing required purpose-built Digital Signal Processing (DSP) hardware. In recent years, the Radisys MRF has delivered best-of-breed IP media processing as a software application deployed on commercial-off-the-shelf (COTS) servers, where the Host and Guest Operating Systems and drivers have been tuned to deliver real-time performance. Most recently, the software MRF technology has evolved to now be deployable as a virtualized MRF in an NFV architecture. Open ization Format vmrf - Package #1 Computing ized MRF Operating System Storage Network Open ization Format vmrf - Package #2 Computing ized MRF Operating System Storage Network ization Layer Intel x86 Xeon 64-bit Architecture Figure 2: ized MRF as two OVF packages running on VMware virtualization technology.

6 6 The Radisys virtualized MRF (vmrf) has been extensively tuned to be deployed as a high-performance, best-of-breed media processing VNF. It is optimized for VMware and other hypervisors to produce high-quality, reliable real-time response in NFV architectures. The Radisys vmrf is designed to deliver optimized performance within an NFV deployment even under high load. Management and Orchestration of the Radisys vmrf Figure 1 showed a simplified view of NFV architecture that omitted Management and Orchestration (MANO) components. MANO components provide critical management facilities for NFVI components and for the set of VNFs deployable on the NFVI. An NFV environment is controlled by MANO functions as defined by ETSI s NFV workgroups. Management and orchestration of MRFs within an NFV-compliant deployment is intended to enable elasticity, managed workflow, and provide QoS for hard real-time IP services managing the entire pool of virtualized infrastructure while maintaining the requirements demanded by a real-time communications system. This orchestration is described in an ETSI NFV case study. OpenStack is an open-source cloud computing software platform often used to provide MANO components in NFV deployments. NFV architecture fits well within the OpenStack framework and its associated functional components. Use cases for MRF management and orchestration include: Onboarding or creating MRF deployable instances via containers such as Open ization Format (OVF). Realized by VMware s vcenter and available through OpenStack Hardware resource management, such as selecting compute servers. Realized by VMware s vcenter and available through OpenStack Instantiating VMs, such as a virtualized MRF on allocated hardware resources. Realized by VMware s vcenter and available through OpenStack Provisioning and configuring virtualized MRFs pre-runtime. Realized by VMware s vcenter and available through OpenStack VMware ization Technology for NFV Rapid deployment and service agility are key value-adds for virtualized infrastructure, as are infrastructure savings resulting from reusability. But VMware knows that you can only fully realize ROI when you consider the whole shared infrastructure, horizontally across the stack hardware, hypervisor, host OS, guest OS, and applications. VMware s features and optimizations, fully supported by the Radisys vmrf, can provide a virtualized environment with real-time media processing performance near that of bare metal. VMware vsphere 5.5, supporting VM hardware version 10, offers specific enhancements to enable fast and predictable response times for latency-sensitive applications with unpredictable workloads. vsphere live migration enabled through vmotion allows the relocation of an entire running virtual machine from one physical server to another, with minimal downtime. In an NFV deployment, this means the vmrf retains its network identity and connections, ensuring a seamless migration process. Management of memory and CPU two resources critical for IP media processing is flexible and configurable at the VM level. Network adaptors have also been optimized for real-time applications. VMware s DRS (Distributed Resource Scheduler) provides an additional layer of flexibility and control of the underlying resources in automated or manual modes to ensure optimal use of the underlying hardware resources as workloads shift overtime and require continual re-assignments and optimizations. Updating configuration post-runtime for service chaining via NFV Forwarding Graphs mechanism Monitoring and reporting on MRF instance performance and resource loading

7 7 OSS/BSS Os-Ma NFV Management and Orchestration Heat Template Service, VNF and Infrastructure Description Se-Ma NFV Orchestrator Heat Engine ized Network Functions (VNFs) Or-Vnfm EMS vhss EMS EMS EMS vcscf vtas vmrf Ve-Vnfm VNF Manager(s) Nova-Compute Cinder/swift Vn-NF Vn-NF Vn-NF Vn-NF Vi-Vnfm Or-Vi Nova Neutron Computing Storage Network ization Layer Vi-Ha Hardware Resources Computing Storage Network Nf-Vi ized Infrastructure Manager(s) = OpenStack Components Network Functions ization Infrastructure (NFVI) Figure 3: NFV architecture with OpenStack component mappings. Instance Fault Management, Reporting, and Corrective Measures. Supported in OpenStack. Turning MRF instances up, down, or re-scaling based on service and business logic. Figure 3 shows a view of NFV architecture mapped to OpenStack components. As a standards-defined IMS core network element, the Radisys vmrf integrates well within OpenStack-defined management and orchestration capabilities. The NFVI Manager manages the complete suite of network elements comprising this infrastructure, including compute, network/io, and storage. The VNF Managers, as well as the Orchestrator functions, enable MRF VNFs to be managed along with other network elements in the signaling domain, such as Application Servers or Session Border Controllers. The VNF Managers run under the control of the higher-level NFV Orchestrator, which is typically responsible for and particular to unique business service logic for example, instantiating MRFs during peak conferencing hours and winding them down and releasing resources during off-hours. The Orchestration controller typically needs to be customizable to individual NFV deployments, as it must provide flexibility suited to the business logic. A deployment within a flexible, NFV-compliant management and orchestration platform such as the OpenStack framework is a cornerstone to the capabilities and performance characteristics expected from a media plane element such as the MRF. Implications for Cloud Providers Media processing quality and performance of interactive 2-way services are realized as a real-time, end-to-end experience, not merely isolated to the QoS performance or capabilities of the virtualized node providing the IP media processing. Metrics and KPIs for end-to-end media processing performance include QoS, bandwidth, response times, redundancy, and failover strategies. The Radisys vmrf can be deployed for private/carrier cloud deployments, such as the AT&T telecom cloud environment, and also in public and over-the top (OTT) services, such as Amazon EC2. Private and carrier clouds offer more opportunities for tuning for example, advanced QoS strategies (eg: QCI controls as described earlier) and MPLS tagging. Public clouds, with proprietary implementations, often align with NFV general principles while not being fully NFVcompliant. Such environments are typically not engineered to provide and support telecom applications and services. Cloud providers need to be cognizant of the hard requirements for IP media processing and engineer the environment accordingly. The Radisys vmrf has been deployed in both private and public clouds. It has been deployed in private clouds adhering to principles of the NFV architecture. The Radisys vmrf has also been deployed as an Amazon Machine Image (AMI) in the Amazon EC2 Cloud.

8 8 Radisys ized Media Processing Solution The telecom industry is in the early stages of a major infrastructure upgrade from network functions running as embedded software on purpose-built hardware platforms to virtualized software functions running on COTS server technology. Many network functions responsible for call orchestration, signaling, and subscriber databases are well suited for early migration to virtualized technology. However, network functions that process actual media packets, such as a Media Resource Function in an IMS architecture, have very stringent requirements for real-time packet processing performance. Hence, the virtualization of an MRF, while certainly possible and proven in telecom cloud networks today, needs to be done with careful technical consideration. The Radisys virtualized MRF (vmrf) has been extensively tuned to be deployed as a high-performance, best-of-breed media processing VNF. The Radisys vmrf implementation for VMware has been extensively optimized to take full advantage Vsphere s real-time performance enhancements and features to deliver high-quality, reliable real-time response in NFV architectures. If you require optimized IMS performance in an NFV deployment, even under high load, a Radisys virtualized MRF with VMware s vsphere, running on modern Intel x86 Xeon 64-bit multi-core processors, provides a robust, reliable, high-performance virtualized media processing solution. References i Available at ii AT&T Domain 2.0 and NFV Architecture: AT&T Vision Alignment Challenge Technology Survey - AT&T Domain 2.0 Vision White 2.0%20Vision%20White%20Paper.pdf iii The latest white paper issued by working members is available here: iv Reference 2: MRF Case Study: MAN001v061-%20management%20and%20orchestration.pdf v From ETSI TS Release 12 (Table 6.1.7) vi GS GS NFV-MAN 001 V0.6.1 ( )) - Appendix (A.1) IMS MRF management and orchestration case study Intel Xeon E v3 Multi-core Processors Breakthrough Performance The latest members of the Intel Xeon processor E5 family deliver the best combination of performance, built-in capabilities, and cost-effectiveness to address technical computing challenges, enable cloud deployments, accelerate processor performance for peak loads, deliver intelligent storage or power data analytics and rapid processing for NFV data flows. About VMware VMware, the global leader in virtualization and cloud infrastructure, enables businesses to thrive in the cloud era by transforming the way they build, deliver and consume information technology resources. Leveraging VMware vsphere, the most widely deployed foundation for cloud computing, VMware enables Enterprises, and Telco Networks to adopt a cloud model that addresses their unique business challenges. About the Intel Internet of Things Solutions Alliance From modular components to market-ready systems, Intel and the 250+ global member companies of the Intel Internet of Things Solutions Alliance provide scalable, interoperable solutions that accelerate deployment of intelligent devices and end-to-tend analytics. Close collaboration with Intel and each other enables Alliance members to innovate with the latest technologies, helping developers deliver first-in-market solutions. Corporate Headquarters 5435 NE Dawson Creek Drive Hillsboro, OR USA Fax Toll-Free: info@radisys.com 2014 Radisys Corporation. Radisys and Trillium are registered trademarks of Radisys Corporation. *All other trademarks are the properties of their respective owners. December 2014