Book Project Why I/O Is Strategic Software- defined Networking Date: April 2013 Author: Bob Laliberte, Senior Analyst What Is Software- defined Networking? The emergence of software- defined networking (SDN) over the past year signals a significant change in network design. Enterprises have heavily virtualized servers and storage, setting the stage for the current expansion of cloud computing environments. Cloud- based data centers let organizations dynamically provision computing power and storage capacity from pools of virtualized resources. But networking technology has lagged behind those developments. Manually intensive network architectures devised years ago for client/server environments now fail to provide optimal agility and underpinnings for the new IT model. SDN, however, aims to create a network architecture that can better keep pace with highly virtualized or cloud environments. It does this by making the traditionally proprietary control plane remotely accessible and modifiable via third- party software clients. This could be a controller or other software. This nascent architecture promises to open a new level of programmability and automation for both network (Layer 2- Layer 3 of the OSI model) and higher level services (Layer 4- Layer 7). The objective: Eliminate the manual processes and configuration inflexibility associated with device- centric networking. Figure 1. Software- defined Networking Source: Enterprise Strategy Group, 2013.
2 Industry is throwing considerable weight behind the SDN trend. Indeed, a number of high- profile vendor announcements and acquisitions make SDN one of the hottest topics in IT. VMware s over $1 billion purchase of Nicira, an SDN and network virtualization specialist, ranks among the most visible developments in the market. Other SDN- related transactions include Brocade s purchase of Vyatta, Juniper Network s acquisition of Contrail, and F5 s acquisition of LineRate. Add to this a number of startups emerging from stealth mode such as Big Switch Networks (which just received an investment from Intel), Midokori, Plexxi, and Pica8. With the spike in industry attention comes the potential for the bandwagon effect or software- defined washing. Some vendors may describe new products or recast existing lines as software- defined even when their actual relevance to the technology may prove quite marginal. Organizations should take the time to understand what SDN is and how it can help them as they explore SDN options. So what networking approaches actually constitute SDN? This architecture pivots on third- party software or controllers that open a network s traditionally proprietary control plane, making it possible to remotely access and modify or program the network (as shown in Figure 1). SDN provides network administrators centralized control, along with the ability to customize highly distributed networks. Conventional networking, in contrast, hinders end- to- end management and limits customization. The proprietary interfaces typical of most networking devices perpetuate a closed environment. Specific technologies associated with SDN include OpenFlow, an open networking foundation specification. This open- standard API links physical and virtual switches (a software based switch functionality loaded into a server or hypervisor) to a centralized controller. OpenFlow is not the only technical underpinning for SDN. Another standard- protocol initiative, the internet routing system, targets routers. In addition, network vendors have developed their own open networks specific to their product APIs, as well as network- virtualization solutions based on virtual- switch environments. In the latter, virtual switches are leveraged to rapidly create logical network tunnels (using VXLAN or NVGRE) and services (virtualized firewalls, load balancers etc.) over existing physical networks. The mix of APIs will continue to evolve. For example, while OpenFlow can serve as the southbound interface between the controller and network devices (both physical and virtual), a comparable standard for the northbound interface has yet to emerge. This will be critical for the adoption of higher level services. Plus, a number of vendors, such as Cisco and Arista, are choosing to create their own APIs to open up access to their switches. In summary, SDN represents a broad network architecture concept that transcends individual APIs. SDN may be thought of as an umbrella term covering physical and virtual networking devices, standard and proprietary APIs, and simplified, centralized automation and orchestration capabilities. Why Do We Need SDN? Several factors and current challenges fuel demand for SDN technology including: Data center consolidation. Organizations report performance and quality- of- service problems as they consolidate data centers and build them out as multi- tenant facilities. Networks are the key bottleneck as they struggle to keep up with the increased traffic loads of larger and more complex computing facilities. Enterprise Strategy Group (ESG) has termed the problem data center networking discontinuity. The dynamic scaling requirements of fast- growing data centers clash with the static, proprietary networking devices that support them. ESG research indicates that more than 60% of enterprises organizations with more than 1,000 employees are either actively consolidating data centers or have completed such projects. 1 Many organizations (48%) will consolidate data centers that formerly supported individual business units into unified, multi- tenant facilities. Such data centers will experience massive scale as they contain more devices, applications, and network traffic. 1 Source: ESG Research Report, Data Center Networking Trends, January 2012.
3 Figure 2. Data Center Consolidation Is your organizaaon currently engaged in - or has it recently completed - a data center consolidaaon project? (Percent of respondents, N=280) No, 37% Yes, 63% Source: Enterprise Strategy Group, 2012. SDN offers a potential solution for the beleaguered data center. The ability to virtualize networks, just as VMware, Microsoft, KVM, and XEN hypervisors virtualized servers, will provide a more flexible environment that more readily accommodates expanding scale and complexity. Widespread virtualization. Server virtualization, meanwhile, also contributes to the need for SDN. Nearly all large organizations are using server virtualization technologies from vendors such as Citrix, Microsoft, and VMware. This IT priority has caused further issues for data center networks. Server virtualization tests networks in a couple of ways. First, this practice increases east- west server traffic, requiring tighter hardware integration and more resources. In particular, many organizations are constrained by 4096 VLANs and need the extra capacity that VXLAN or NVGRE can deliver (up to 16 million). Second, running a number of virtual machines on a single piece of hardware and enabling automated live- migration and provisioning of resources quickly diminishes visibility, and complicates troubleshooting. By deploying DN architecture that control both the physical environment and the virtualized network environments, organizations can create high performing virtualized network environments that enable them to rapidly scale, and change network configurations and services to map to highly mobile VMs. This process eliminates manual reconfiguration, reduces costs and errors, and accelerates problem resolution. Growth of private clouds. And, naturally, the enterprise pursuit of private cloud computing, which builds upon mature virtualized environments to add higher levels of automation and makes it easier to self- provision services, places additional stress on networks to respond to the changes and raises the profile of SDN as a potential remedy. Taken together, private clouds and virtualization represent a high IT priority, which will probably drive demand for SDN. Sixty- three percent of the respondents to a recent ESG survey planned to increase private cloud infrastructure and virtualization software spending in 2012. 2 Similar to the previous bullet, as virtualized environments mature into private clouds, the network needs to become much more flexible 2 Source: ESG Research Report, 2012 IT Spending Intentions Survey, January 2012.
4 and agile. SDNs hold significant promise to enable not only flexibility for the physical and virtual network, but are also the foundation to deliver higher levels of automation and orchestration that are required in a cloud environment. Myriad manual processes. Another reason that organizations now consider SDNs: the abundance of manual processes involved in maintaining a legacy network. Enterprise network teams responding to an ESG survey identified an excess of manual processes and difficulties/delays associated with change and/or configuration management as network operations problems that they have experienced, making them the top two responses to the question. 3 Against that backdrop, interest in a new network architecture seems hardly surprising. SDN s centralized control plane streamlines operations by providing network administrators the ability to automate an integrated network fabric. For example, to execute a change in the network configuration, an administrator needs only to make the change once in the controller GUI in order for it to propagate to all the network devices under its control. This approach eases the management burden. In a legacy network, each network device has its own control plane and must be individually configured. This requires time, people, and a repetitive manual process that could be prone to errors. Staffing Demands. Centralized management and automation also address the staffing demands of running a network. Organizations facing the need to hire more administrators as data centers expand may be able to maintain staffing levels with an SDN architecture. In some cases, the efficiency gains may let IT shops reassign administrators formerly dedicated to configuration duties to high- value projects. Having a centralized control plane should enable organizations to manage larger environments with the same number of staff. As the example in the previous bullet explains, just making a simple configuration change becomes a one person task with a few keystrokes, vs. an all- day event. A network that is easier to manage provides an added benefit: responsiveness. That characteristic becomes particularly important in dynamic, self- service cloud environments, which call for rapid service delivery. Flexible, programmable SDNs will outperform traditional networks in this setting. Essentially, the promise of SDN is to enable organizations networks to keep pace with highly virtualized server and storage environments and eliminate a bottleneck for provisioning new applications and services. Legacy networks utilize an inefficient node- by- node management paradigm, in which manual processes make moves, adds, and changes time consuming and inefficient. All these challenges manifest themselves as additional cost to the business, in the form of additional hardware, software, staff, and lost business opportunities. Comprehensive SDN deployments should be able to reduce costs, optimize the virtualized environments they support, and accelerate IT s responsiveness to the business. Who Will Adopt SDN? Initially, the largest computing facilities with highest growth demands are the main prospects for SDN deployment; however, over time this should be universally accepted technology. Cloud service providers are among the early adopters. Those companies operate massive data centers so networking is core to their businesses, and they tend to grapple with scalability issues. Google, for example, is one of the early adopters of SDN and OpenFlow. In fact, Google was among the six companies that launched the SDN- promoting Open Networking Foundation in 2011. The other founders are Deutsche Telecom, Facebook, Microsoft, Verizon, and Yahoo. Academic and research institutions also populate the list of early adopters. The OpenFlow- based SDN approach stems from a research effort involving Stanford University and the University of California at Berkeley. Indiana State University is also a major proponent and user of the technology as well. 3 Source: ESG Research Report, Data Center Networking Trends, January 2012.
5 On the network virtualization side of SDN, organizations like NTT and Ebay have reported initial success in virtualizing their network and dramatically reducing provisioning times as well as enabling widespread VM migrations. The early adopters will likely pave the way for a more general uptake of the technology, perhaps similar to the way Hadoop and other big data systems migrated from specialized use cases to more mainstream acceptance. As SDN technology and standards mature and move more broadly into corporate enterprises, one can expect those organizations operating enormous, service provider- like data centers to ride the initial wave of adoption. Additionally, data center consolidation and increased private cloud deployments will inspire more enterprise IT departments to consider SDN. In general, SDN offers a good fit for organizations with exceedingly large and complex networks dealing with growth and scale. Why Is I/O Strategic to SDN? SDN places data center networking the last data center component yet to be heavily virtualized on a par with server and storage virtualization. As part of the network ecosystem, the I/O system, however, also has the ability to play an important part in a software defined network. The I/O system presents a new area of programmability and organizations need to think about how they can leverage I/O systems and architectures to deliver services and improve on end- to- end performance. Figure 3 illustrates how an I/O system, already in the path between the virtual and physical switches, could be included in an SDN environment. Figure 3. I/O Systems as an Area to be Programmed Source: Enterprise Strategy Group, 2013. However, with an intelligent I/O system deployed at the edge, organizations could leverage the distributed edge processing to enable more effective I/O and network virtualization. By taking advantage of Single Root I/O Virtualization
6 (SR- IOV) and support for VXLAN and NVGRE, the I/O system architectures have transformed into intelligent network edge devices and can assume the network processing previously done on the virtual switch, freeing up host CPU. The implications of this shift are clear: Increased CPU processing on the host enables higher VM densities, which reduces the number of physical hosts that are required and lowers both CAPEX and OPEX. This does not mean the virtual switch is eliminated rather it becomes a lightweight control plane that takes advantage of the intelligent I/O system edge processing power. In these instances, the diagram may look more like Figure 4. Figure 4. Intelligent I/O Systems Offload Virtual Switching Functions Source: Enterprise Strategy Group, 2013. Just as the physical and virtual network devices are being programmed, the ability to leverage intelligent edge I/O systems could prove to be very useful for cloud environments, especially with greater levels of east- west traffic between servers and more mixed workloads in production environments. The intelligent edge systems will be able to utilize virtual edge bridging (VEB) to redirect VM to VM traffic without the need for a virtual switch or without incurring any negative impact to the CPU cycles. As these technologies continue to progress, expect this technology to make its way into OpenStack orchestration releases and the Quantum network component. Cloud service providers should welcome these advances as it will enable them to drive higher VM densities per compute device and provides an opportunity to inject services in the I/O path. This could have a significant positive impact on the network virtualization portion of SDN. Think about network virtualization as taking all the physical network devices and making them appear as one big switch. The ability to quickly allocate and re- allocate networks as workloads spin up, spin down, and move will be imperative in a highly virtualized environment. Network virtualization technology enables organizations to rapidly create virtual network links (using VXLAN or NVGRE) from one point to another. Currently, organizations rely on virtual switches to handle this network processing. However, intelligent I/O systems with edge processing could accomplish that task.
7 As network capacities increase in scale to 10, 40, and eventually 100 Gb/sec, the ability to accurately control and apply programmability at the I/O system level will become more important. Greater visibility into the traffic and ability to inject services from a central control plane will help to drive efficiencies and reduce errors. According to ESG research, the transition to higher capacity links are well underway with 67% of enterprises reporting that they are actively moving to 10GbE or have already set 10GbE as the standard. 4 In fact, some organizations are pressing 10GbE into service to support their cloud initiatives as nearly half (48%) of the large organizations with more than ten data centers surveyed told ESG that they are adopting 10GbE for current or future implementation of a private cloud in their data centers. 5 These cloud environments will require higher levels of automation and programmability, and having intelligent and programmable I/O systems at the edge will enable these organizations to transition to an SDN- enabled cloud environment without any drain on the CPU. Given these advancements, organizations should strongly consider taking an additional step to include intelligent I/O systems as part of their SDN environments. Service providers and large enterprises with complex environments should begin to investigate how intelligent I/O systems can enhance an SDN architecture. The Bigger Truth SDN is poised to make a significant impact on large data centers, offering an entirely new take on network architecture. The key advantages of SDN include centralized control, programmability, and the automation of change and configuration tasks. To take full advantage of those benefits, SDN deployments should encompass unified, end- to- end network fabrics, including intelligent I/O systems. Intelligent I/O systems have the capability to deliver significant advantages in an SDN environment and especially in those looking to leverage network virtualization capabilities. Its ability to perform I/O and network processing independently from the host CPU enable higher VM densities, fewer physical hosts, and lower CAPEX and OPEX. This should prove to be very beneficial to cloud providers trying to squeeze out every last bit of efficiency from their environment and enterprises looking to accelerate private cloud deployments. The intelligent I/O systems with edge processing present an opportunity to inject programmability at the edge and positively impact host CPU performance and existing server virtualization technology. So while most of the SDN buzz has been focused on physical switches, virtual switches, and controllers, architects would be well served to investigate the potential benefits of intelligent I/O systems capable of network processing to strengthen their SDN deployments. All trademark names are property of their respective companies. Information contained in this publication has been obtained by sources The Enterprise Strategy Group (ESG) considers to be reliable but is not warranted by ESG. This publication may contain opinions of ESG, which are subject to change from time to time. This publication is copyrighted by The Enterprise Strategy Group, Inc. Any reproduction or redistribution of this publication, in whole or in part, whether in hard- copy format, electronically, or otherwise to persons not authorized to receive it, without the express consent of the Enterprise Strategy Group, Inc., is in violation of U.S. copyright law and will be subject to an action for civil damages and, if applicable, criminal prosecution. Should you have any questions, please contact ESG Client Relations at 508.482.0188. 4 Source: ESG Research Report, Data Center Networking Trends, January 2012. 5 Source: Ibid.