5G is a Delicate Balance
Part 3: In the white box era, there’s still room for dedicated equipment
Welcome back to my blog series: ‘5G is a delicate balance between…’ where we’ll discuss the many aspects affecting the evolution to 5G networks over the next few years. We’ve already covered the investments v. ROI of 5G, and explored smart edge versus intelligent core transport. In this third part of my series, it’s time to address the balance between white boxes and dedicated equipment.
This year, AT&T announced plans to deploy over 60,000 ‘white box’ routers across its wireless access network in the US, promising to lower costs and speed the carrier’s path to 5G. The company says that it believes a “virtualized, open-source, white box approach is the best way to go,” but is it? While the trends of virtualization and disaggregation seem to be at their peak, in my humble opinion, there is still a need for dedicated networking equipment.
Problems with the box
Networking equipment has traditionally utilized dedicated ASICs to provide the most optimized solutions. But as we know, dedicated equipment comes with many of the downsides that were highlighted during the SDN and NFV era. ‘Softwarization’ was hailed as the savior, and cost cutting was to be expected by removing the ‘hardware’ from the equation. As the world was moved by digitalization, virtualization and disaggregation, we network vendors were forced to rethink our portfolios and solutions. Now it seems that many vendors have both stand-alone, dedicated equipment and a software version that can run on any ‘white box’ (e.g. COTS platforms).
As I’ve discussed in previous blogs, however, that little ‘white box’ comes with plenty of issues. Here are just a few:
- Price/TCO: VNFs do not necessarily have lower TCO than that of a purpose-built device (if we take into account: the platform cost, higher power consumption, and SW licensing), and in some cases, they are either comparable in price or more expensive.
- Performance: In some cases – running the same software on a white box – as opposed to dedicated equipment –doesn’t work as well. So while you think you’re saving money, you’re actually getting lower performance and spending more to make up for it in the long run.
- Efficiency: Because the COTS platform is much less efficient than ASICs in dealing with data plane I/O, an intensive part of the switching and routing process needed for networking equipment.
But does that mean we should ‘throw the baby out with the bathwater’? No, there are benefits to virtualization:
- Greater agility: Software is much easier to update and doesn’t require those dreaded truck-rolls to exchange/upgrade physical hardware for updates every time
- Space savings: Virtualization enables you to squeeze many functions onto a single HW platform. If each VNF runs only occasionally or part of the time, then the resource sharing enabled by virtualization may compensate for the lower efficiencies described above
- Simplify operations: Enabled by remotely rolling out new versions, making sure applications are up to date, the endless possibilities of adding/changing/replacing software at the click of a button. Moreover, in many cases customers can also reduce the total number of ‘solutions’ vendors turning the service providers into their center of support
- Flexibility and openness: As all of the functions are SW based, it is very easy to introduce new and innovative VNFs, combine them with existing VNFs, replacing existing VNFs and have a thriving “app store” ecosystem
- Better pricing models: SW based VNFs enable flexible licensing with “pay as you grow”, “pay as you use”, and subscription based models that are more economical (by deferring CapEx to OpEx) than those of dedicated HW (that must be bought upfront)
Balancing Hardware and Software
5G will expect networks to perform at levels never seen before, but that doesn’t necessarily mean deploying the shiniest, newest solution every time. Our next ‘G’ absolutely requires a balance between the dedicated equipment of yesterday and today’s white boxes.
We’re starting to get a better understanding now of where white boxes can be the most use and worth the implementation – and where they may not be the best fit. There are some cases in which programmable software is not as good as a dedicated appliance for it because there are several factors that need to be considered:
- Needs and functionality – Fixed functionality and upfront needs favor ASIC based solutions that are task optimized
- Where in the network you want to deploy – at the edge, where diverse and dynamic functionality is required (and thus more suited to NFV) or core, where the functionality is much less diverse and much more static
- Utilization: Lightly loaded vs fully loaded NF - Lightly loaded NFs (which are more typical at the edges of the network) can be “squeezed” together and thus be more cost optimized. The core has much higher load factor, due to the law of big numbers and thus more optimized for ASICs
- CPU vs I/O cycles – The higher the ratio, VNFs are better solutions. Core routing and switching are better done with ASICs, and this will not change in the foreseeable future
As things stand today, price-performance still favors dedicated equipment for many network applications. White boxes have started to implement some CPE functions, like basic routers and firewalls, but this doesn’t yet cut it for processing-intensive functions.
Any platform deployed today needs to be future-poof and able to provide the capability and flexibility customers need. Not just now, but in the future. Understanding the tradeoffs and capabilities of each technology, as detailed above, will enable the CSPs to take more informed, future proof decisions.