Though much has been said and written about SDN (software defined networking), little has been said about its business case. We know at a high level that the service provider board members of the Open Networking Foundation including Deutsche Telekom ( DTE.DE ), Facebook, Google ( Nasdaq: GOOG ), and Verizon ( NYSE: VZ ) view SDN as a means to reduce their costs and increase service delivery velocity. A more detailed view of the SDN business case can be developed by digging into the cost structures of these large service providers.
The SDN business case must necessarily emphasize operating rather than capital expense reduction. Large service providers spend much more on operations than capital equipment. Google, for example, spends $17 on operating expenses for every $1 of new equipment while Verizon spend $6 of expenses for every dollar of equipment, according to their latest financial statements. One justification for SDN is that it will allow service providers to switch from proprietary equipment to commodity off-the-shelf equipment (COTS). My own studies show that SDN does allow use of switch/router line cards with less onboard processing power. This reduces the cost of routers by 20 percent to 50 percent; however, routers at most account for about 5 percent of a service provider's capex, and thus would not provide the big change in cost structure that service providers seek.
Service order processing, however, accounts for a big piece of operating expenses and is dramatically affected by SDN. Service order processing workflow involves many operations support systems (OSS), customer facing and network facing activities. These activities include call center operations, service ordering and provisioning, and account management and billing. Under the present mode of operations (PMO) customer facing interfaces are maintained between each customer's activity and its associated OSS component, and network facing interfaces are maintained between each OSS component and element management systems associated with each network element. Under the PMO many disparate databases, automated and manual processes provide the linkages among customers and the underlying network elements (network infrastructure). SDN makes the linkages among customers and network elements explicit and automated. Databases are standardized and consolidated. All APIs are OpenFlow. REST software architecture is used throughout rather than various vendors' proprietary EMS systems, and all manual interfaces are eliminated. This simplifies service order processing and results in lower setup and processing costs and higher service velocity. SDN reduces the cost and speed of service order processing by about a factor of seven as compared to the PMO. This processing cost reduction combined with the preponderance of operations expense over capital expense accounts for the majority of SDN's business benefit. A seven-times increase in service velocity also yields important business flexibility and responsiveness benefits.
Specific cost reduction results for several promising use cases are discussed in the following paragraphs.
Google was the first, to my knowledge, to implement data center network virtualization--also called hyper-scale data center. The concept is to extend the compute and storage virtualization concept across multiple data centers using a virtualized WAN. This allows simultaneous optimization of the location of VM loads and traffic routing on the WAN. For example, whereas traffic engineering seeks to find the lowest cost route across a network to connect VMs to an end-user, network virtualization can find an even lower cost solution by also reassigning the VM load to a data center that is closer to the user. This joint optimization of network capacity and the location of VM loads can potentially reduce data center total cost of ownership (TCO) by as much as 36 percent by reducing the peak capacities of all data centers in the network. Also, the WAN's optical transport costs can be reduced by 25 percent. These savings are in addition to the service order processing cost reduction discussed previously.
SDN service creation and insertion, and network analytics use cases, employ an SDN-enabled switch/router to direct traffic flows into and out of farms of security, traffic management, and monitoring tools. Security tools include firewalls and intrusion protection systems; Application Delivery Controller is a traffic management tool; and and VoIP analyzers are network analytics tools. The SDN-enabled switch enables customer-specific policy-based service delivery and enforcement. SDN links the network infrastructure to the customer-facing applications. This linkage reduces the amount of traffic that flows through the service creation and network analytics tool farms, and thus improves capacity utilization of the individual tools/appliances. The big strategic benefit, however, comes from faster service creation and real-time network statistics collection and alerting.
The campus network slicing or bandwidth-on-demand use case has been implemented on several campuses of large research-oriented universities. It provides a virtual network with security policies and bandwidth guarantees overlaid on shared network infrastructure. These networks are used by networking intensive academic departments such as physics and meteorology. As is the case for service provider networks, the big business benefit is simplification and cost reduction of the IT department's management processes. An IP/MPLS network is another way to provide separate virtual networks on common networking infrastructure. This approach involves distributing the Layer 2/3 control plane function onto firmware in each switch/router. I have compared this approach with a centralized SDN controller and found that the TCO over five years of the SDN controller is about 30 percent lower than that of the IP/MPLS solution.
Major content- and network-focused service providers established the Open Networking Foundation as a vehicle for dramatically reducing their costs of service delivery through open software defined networking technology. Though SDN requires rethinking the design of L2/L3 networks, the SDN business case must be anchored on getting the cost out of network operations. Protocols such as OpenFlow and software such as REST hold the potential to link business applications to network infrastructure as a means of making near-order-of-magnitude cuts in service providers' operating expenses.
Michael Kennedy is a regular FierceTelecom columnist and is Principal Analyst at ACG Research . He can be reached at mkennedy@acgresearch.net.
No comments:
Post a Comment