Looking at the development of mobile video and the IoT, the service provider market is changing dramatically. Margins for their traditional services are dwindling while many new and potentially lucrative revenue opportunities are waiting in the starting blocks. To meet the challenges, service providers (SPs) all over the world have to determine how to best migrate to the highly agile, highly efficient virtualized/software-defined network architectures which they will need to successfully compete. Choosing the right COTS server hardware at this moment is no simple matter. In fact, the right hardware choice will significantly impact the ability of SPs to maximize operating margins by both better responding to customer desires and reducing total opex. The wrong hardware, on the other hand, will hurt profits by impeding responsiveness to market opportunities and driving up costs.
SPs have to quickly overhaul their service delivery infrastructure to profit from the new market opportunities and such overhauls must also position them to capitalize on longer-term market opportunities which may be difficult to fully forecast. By now, this is best done through software-defined environments that enable or are complementary to network functions virtualization (NFV). With this SP’s value-added capabilities are uncoupled from the underlying hardware, so that compute, network and storage capacity can be flexibly allocated wherever it makes the most sense – and so that SPs don’t have to make burdensome investments every time they have to implement a new capability or enhance an existing one.
Careful selection of the COTS hardware
Given the virtualization of these higher-level functions, SPs can refresh their infrastructure with commercial off-the-shelf (COTS) server hardware. The move to COTS hardware will allow SPs to minimize initial hardware refresh investments and to inexpensively add incremental hardware capacity only when it is absolutely necessary. SP infrastructure planners should therefore not make any hurried decisions in selecting the COTS hardware fabric on which they begin to build their new virtualized environments. The careful consideration of the technical, operational and economic requirements of SDN/NFV will likely bring out five key attributes of NFV-optimized server infrastructure:
The right functional attributes
The right functional attributes enable SPs to respond more adaptively to even the most sudden and unexpected shifts in customer demand. Next-generation NFV environments will require modularized hardware that combines compute, network and storage in a single, low-TCO unit. On the compute side, SPs will likely need some variability in their configurations in order to accommodate different workloads. However, those variations should be provided within a common “package” across all nodes to keep the overall service delivery environment as simple as possible. On the network side, each converged hardware unit should provide redundant 1GbE or 10GbE connectivity, as well as an option for load balancing functions to ensure optimized utilization of available on-board resources. Ideally, switching functions will support the OpenFlow protocol, which is a primary enabler of software-define networking (SDN).
The right performance attributes
The right performance attributes enable SPs to optimize delivery at scale for video, IoT and/or other emerging services that drive profits, market-share and brand value. Some SP infrastructure planners may think that they are able to compensate for under-performing COTS hardware by simply deploying more of it – but that will get too expensive in the long run. SPs need to be able to acquire the performance they need in small enough increments to avoid large “step function” capex outlays and to align allocation of available capacity to multiple diverse workloads. And this is what makes the right balance essential. A common way to quantify infrastructure performance is by processor speed, memory I/O and other technical attributes. And in addition to comparing raw hardware specs, SP infrastructure planners should carefully evaluate benchmarks for how alternative converged infrastructure solutions handle anticipated real-world workloads. Such benchmarks may include video channel density (e.g. number of simultaneous live HD video streams support-ed or real-time VoD offline transcodings per file) and/or IoT/Mscalability (e.g. number of CoAP requests per second), and real-time Hadoop based data analytics.
The right partner ecosystem:
In the best case, SPs are able to most freely choose best-in-class software partners that can help drive competitive advantage. Limitation can be as much of a constraint on a SP’s ability to quickly and efficiently meet market demand as the speeds and feed of the hardware itself. The most developer-friendly software designs are those that conform to industry standards (OpenStack, OpenFlow, etc.) and that leverage widely embraced technology. In addition it’s always a good idea to “shortlist” vendors that have already certified NFV software partners with solutions that align well with near-term needs. And last but not least: SPs should only give their business to NFV hardware vendors with programs and processes in place to accelerate certification of innovative software developers on their platforms.
The right physical attributes:
The right physical attributes aggressively drive down costs at the core and the edge. Core density is obviously a central consideration, but also other hardware properties may affect physical management of the data center. SP infrastructure planners should be careful about giving space and power consumption the highest priority. Microservers, for example, offer an attractive low power draw—but are typically “application-defined” only for web an online services, and are poor candidates for NFV infrastructure given their nonstandard architecture larger footprint, limited networking and storage capabilities and inability to segment management data and traffic data.
Proven telecom quality:
Virtualization certainly does provide some degree of failover, but it does not render SPs immune from hardware failures. Nor does it give them leeway to constantly take hardware out of service for scheduled maintenance operations. In fact, the economics of virtualization make sense by intensifying per-unit workloads. SP infrastructure planners should therefore look for COTS hard ware with carrier-class characteristics including hot-swappable modular servers, redundant chassis fans and tool-less FRUs. They should also look for COTS hardware vendors that have demonstrated the ability to deliver products that are fully supported during a useful life of 5-7 years, rather than the shorter life spans associated with cheaper enterprise-grade hardware.
You can obtain the full COTS-NFV overview by downloading the Kontron-sponsored Morgan Armstrong Whitepaper “Big Picture, little Picture” by visiting http://www.kontron.com/biglittle.
Did you already refresh your infrastructure with commercial off-the-shelf (COTS) hardware? Tell us about your decision criteria by leaving a comment.