 | Issue: | Europe II 2014 |
| Article no.: | 14 | |
| Topic: | Software-Defined Networks and Network Functions Virtualization Fuel a Cloud Computing revolution | |
| Author: | Jarrod Siket | |
| Title: | SVP & GM, Marketing | |
| Organisation: | Netronome | |
| PDF size: | 280KB |
About author
Jarrod Siket is Netronome’s Senior Vice President and General Manager, Marketing.
Jarrod has 20 years of IP networking experience, including product management, strategic planning, sales, business development and executive management and is responsible Netronome’s processor, platform and software products.
Prior to Netronome, Jarrod was VP of Marketing at Tollgrade Communications, focused on carrier IP network service assurance. Jarrod also spent ten years at FORE Systems, acquired by Marconi Communications, as a senior director of product management and, later strategic planning, where he also held several senior positions in industry standards bodies related to IP, MPLS and Ethernet networking.
Jarrod holds an MBA from the Katz Graduate School of Business at the University of Pittsburgh and a BS from Carnegie Mellon University.
Article abstract
SDN and NFV are not only changing networks architecture, but are also the design of the devices that make networks. The trend is away from proprietary servers, towards dynamically switched virtual ones, to simplify deployments, extend hardware lifetimes and minimize management costs. Cloud hyperscale data centers require such automation and orchestration to cope with the huge numbers of switches and servers that are needed to satisfy the escalating demand. However, the increasing workload of midware SDN servers, which now require to host core applications, network address translation, tunnel origination and termination, cryptography, QoS and metering, brings low performance. To alleviate these new bottlenecks, the server’s performance can be accelerated by selectively offloading functions, under the standards of NFV. Acceleration cards can be used to bridge between legacy and SDN servers, and offload tasks for a wide range of services, thus dramatically reduce costs.
Full Article
Networks and the applications and services they deliver are undergoing a once-in-a-generation change. Cloud computing is driving a new cost-effective and scalable model for outsourced applications and services, as well as on-demand access to network and compute resources. This vision promises to revolutionize every aspect of communications from the pace with which innovative applications are developed, to reducing the time, cost and complexity of building, provisioning and managing the networks that support them.
These benefits can only be realized through changes to the way networks are architected and designed, including changes to the products that are used to build them. Many network architects and operators believe that emerging technologies such as server and network virtualization, centralized orchestration, policy control via software-defined networking (SDN) and network functions virtualization (NFV), represent the best ways to realize this vision.
Cloud networking emerges
Public and private clouds represent the shift of applications and services out of enterprise networks and into shared network locations. Hybrid clouds provide enhanced capabilities to partition resources across both public and private domains as well as the ability to move instantiations of these applications between the clouds.
The computing resources made available are now virtual machines on standard servers within the cloud, that serve as hosts for proprietary business applications, standard business applications (such as Hadoop, MapReduce, Apache, and MySQL), and advanced services such as cybersecurity and network analytics.
A new network architecture is required to support the shift of these massive compute resources from private enterprise networks to the cloud, as well as the technical requirements behind the ability to quickly add, modify and move resources within networks.
Standard compute versus custom closed systems
The first major change is the shift from proprietary, closed networking systems based on custom hardware and software. Hyperscale data centers, as well as rapidly following enterprise and carrier networks, are now deploying software-controlled, commodity systems. This is not only happening in the traditional server locations but also in places once reserved for custom platforms, such as switching, routing and advanced services. This change aims to rapidly increase innovation, reduce capital costs, simplify deployments, extend hardware lifetimes and minimize management expense.
Hyperscale data center networking architecture
The underlying network architecture for a cloud data center is unique relative to legacy network designs. These new networks must now support thousands of switches and routers that make up a high-bandwidth Ethernet and IP infrastructure capable of providing connectivity to hundreds-of-thousands or millions of commodity servers, each hosting hundreds of virtual machine instances.
To manage a network of this immense scale, new technologies are required for provisioning and operation. Global resource orchestration, enabled by software platforms such as OpenStack, provide the ability to provision virtual machines, assign applications and services to them, migrate them as needed, associate them with a unique tenant network space, and create the connectivity between them via a virtual overlay networking. Centralized policy control that defines the communication paths for the virtual overlay networks is also required and is provided though protocols such as OpenFlow, defined by the Open Networking Foundation.
Intelligence moves to the edge
A hallmark of this new network architecture is the shift of intelligence to the network edge. In legacy networks the attached servers are end-points connected to a network that is responsible for the raw connectivity, quality of service, reliable forwarding, and advanced services such as security, load balancing and analytics. The new architecture features a simplified core, surrounded by highly intelligent edge servers that are an equal part of the network topology, and now responsible for the advanced services and forwarding decisions.
In addition to hosting the core applications, the servers are now tasked with network and security workloads that include: network address translation, tunnel origination and termination, cryptography and other security processing, load balancing, QoS and metering. Software-based virtual switches are responsible for implementing these functions within the system.
The high costs of virtual networking
While this new architecture could meet the requirements of the cloud-based, hyperscale data center, it quickly became evident that the single greatest risk to its scalability was the increased burden and the associated reduced performance of the virtualized servers. Even within the smallest networks, workloads in the software-based virtual switches within each server had a significant impact on the overall performance. The virtual switch is responsible for all of the packet processing, tunnel encapsulations, load balancing, memory copies, flow table management, access control lists, filtering, and distributed security functions such as firewalls. These tasks consume valuable resources from the general purpose CPU, restricting the network performance of the system as well as the core applications that are housed with the virtual machines on each server.
Intelligent offload of virtual switches
The networking OEMs, server ODMs, and large-scale enterprises at the forefront of these network changes have been the first to encounter these scalability bottlenecks. Many have chosen to accelerate the server’s performance by selectively offloading the virtual switch. This is achieved by using high-performance network interface cards in standard servers that are optimized to handle the most burdensome compute workloads of the virtual switch, and return the valuable compute resources to the server and its applications. This virtual switch acceleration removes and I/O bottlenecks from the servers, eliminates the high overhead from virtual switching, improves server connectivity with greater port density and bandwidth, provides feature parity among the hardware and software virtual switches in the network, and maintains the software-defined networking control and orchestration.
European telecommunications standards institute: Network functions virtualization
Perhaps the greatest benefit of using accelerated standard server platforms for high-performance networking designs, is the presence of resident general purpose computing resources. Network operators are burdened with the cost, space, power and management complexity of deploying closed, proprietary systems for individual applications and services. The Network Functions Virtualization specification created by ETSI has a stated goal “to address these problems by evolving standard IT virtualization technology to consolidate many network equipment types onto industry standard high volume servers, switches and storage. It involves implementing network functions in software that can run on a range of industry standard server hardware, and that can be moved to, or instantiated in, various locations in the network as required, without the need to install new equipment.”
A new class of networking devices
With this new design, the virtualized servers participating in the overlay networks can now fully realize the vision set forth by SDN and NFV. More importantly, the accelerated server design enables several new types of devices that further support the hyperscale data center networking model:
• SDN application gateways: Software-defined networks have emerged in pockets, and require connectivity to legacy IP infrastructure. The gateways required to bridge these two network types can now align with the SDN and NFV visions. Standard servers with network acceleration cards can be used to handle tunnel creation and translation, combined with various co-resident applications and services, all under centralized OpenFlow control.
• SDN and NFV middle-boxes: Legacy networks are littered with application-specific, close, proprietary hardware and software devices that deliver cyber security, load balancing, analytics and other network services. As a result of this new architecture, standard servers with acceleration cards that offload the virtual switching can serve as shared host for a wide range of services, dramatically reducing costs.
• Intelligent top of rack switches: Legacy data center switches are based on custom hardware and provide basic connectivity between rack-mounted servers and the network backbone. New intelligent top of rack switches can also be based on standard server platforms that provide the same connectivity but also general purpose computing resources to provide distributed network and security services for traffic to, from and among racks.
• Software-defined servers: The same network acceleration cards can be deployed in every standard server within the hyperscale data center, significantly improving network performance and returning valuable compute resources to the general purpose CPU and their associated applications.
SDN and NFV are not only changing the way people are building networks, but are also changing the design of the devices that make networks. This has a direct impact on the long-standing ecosystem of suppliers of communications equipment as well as the hardware and software technologies that define them. The net results are better for users, their application performance, and costs of operating the network. As a consequence of these changes, some long-standing companies may fail to react and respond to these irreversible changes, making room for a new generation of innovative start-up companies to emerge.
