In this series on network functions virtualization (NFV), we’ve been spending time talking about some of the pitfalls and challenges involved in adopting NFVI components – and how to plan for deployment. This week we’re going to spend some time breaking down what to look for in moving to NFV hardware components.

The information I have gathered is based on feedback from various service providers and product experts on how to get NFVI ready for prime time. They have identified many of the key areas to look for in NFV hardware components.

As I’ve pointed out in past articles, telco clouds are different animals, so they are going to have key differences from your average IT cloud system.

There is an essential difference between a typical cloud platform and a telco cloud platform using NFVI. Workloads in the telco cloud often require deterministic and dedicated allocation of resources, according to Javier Gavilán, planning and technology director and global CTO at Telefónica. This means that in many cases, to be carrier-grade, there will be no resource sharing. This is different from the usual cloud computing practice, where resources are shared and even allocated with over subscription.

Thus, says Gavilán, “This requires careful selection of hardware components, so the network virtualization environment will be configured differently from a cloud environment. The purchase of both technologies may follow the same process – the space must be shared – but hardware and software are different. All are x86 architecture, but mounted in a different way.”

Michael Kozlowski, vice president of product management at Windstream, says that any “carrier grade” NFVI must address performance, programmability, security, reliability, and agility – but it also must be fully programmable. To achieve this, he believes, the hardware components, like the software components, should be disaggregated.

To achieve these goals, setting up a carrier cloud platform requires careful selection and guidelines for each component. Some of the considerations are:

• Which cloud management system is going to be used, and what are the specific hardware requirements? For example, OpenStack has many different distributions – once a distribution is selected, the requirements will have to be matched with the hardware.
• What type of servers will be needed? Different servers in the NFVI cloud will have different requirements. For example – is it a deployment server, a control server, or a compute server? What are the characteristics of each, and which requirements will it have for computing power and memory?
• What are the storage needs? Many different applications have different storage needs. Are hard drives good enough, or does the NFVI system require high-performance (and more expensive) solid-state drive (SSD) systems?
• Are there heavy I/O demands? As we’ve identified many times, telco clouds often have intense I/O processes that may require specialized hardware such as network interface cards (NICs) with software and hardware offload features.

As each carrier plans its NFV hardware strategy, there are a number of resources to consult to set up best-practices and see how these systems are designed. For example, Verizon has published a detailed plan of its SDN-NFV reference architecture, which outlines some of the key requirements for NFV hardware. These requirements include:

• Modular and extensible hardware that shares communication fabrics, power supplies, cooling units, and enclosures
• Redundant and highly available components with no single point of failure for
  1. Communication fabric (NICs and enclosure interconnect)
  2. Power supplies, cooling units/fans, and management processors
• Support for a non-blocking communication fabric configuration
• Out-of-band management
• Support for advance network I/O capabilities, such as single-root I/O virtualization (SR-IOV) and data plane development kit (DPDK)
• Plug-in card support for workload-specific items such as
  1. Compression acceleration
  2. Media-specific compute instructions
  3. Transcoding acceleration
  4. Graphics processing

As you can see, there are many variables – and each NFVI installation will be different, depending on the specific situation and requirements of the customer.