Grant Lenahan, Partner and Principal Analyst
We’re on our way to virtualization – primarily NFV and SDN, but where are we in terms of real benefits? Are we on track? Where do we need to focus? Appledore Research Group spoke to several Tier-1 CSPs and suppliers (more to come) and so far they have validated both our hopes and our concerns.
We believe that major opex, capex and agility benefits are possible, and that automation is absolutely essential. So far that hope has been confirmed – CSP leadership is on board.
We also believe that the journey, and especially the implementation of management platforms, is only partially complete, with several limiting factors that we will itemize below. That concern has also been validated (more by CSPs than by suppliers, naturally).
Summary: we are on the right track, but still far from the point where most operational, performance and capital utilization benefits accrue.
Over the past month or so I have had the privilege to speak with senior technology executives at leading CSPs. I’ve also been privy to many discussions from suppliers that were documenting their progress – on everything from existing product lines to how to view AT&T’s ECOMP open-source bombshell.
Let’s use ECOMP as a jumping off point. AT&T clearly calls ECOMP a “platform for automation”. That’s meaningful; its not an “NFV” platform or a “virtualization” platform, nor is it the dreaded “next gen” platform. They stated their goal: automation. This is the right goal; automation means agility, cost reductions and overall business competitiveness.
To make the business-case work, we need to following:
- Self-realizing NFVs and services, based on parametric model-driven orchestration. Pre-defined workflows don’t cut it
- Micro-services based NFVs that can be easily and incrementally healed and scaled
- HA and scaling implemented by the cloud platform, not built into each NFV
- A large and ubiquitous pool of NFV-I – within reason, any NFV should be able to run on any node, in any datacenter, from any vendor
- NFVs that perform similarly to their appliance based predecessors. While beyond the scope of this Blog, it is likely that we will need infra-aware MANO, specialized packet and radio processing accelerators, and other performance enhancing technologies.
Our informal but often deep conversations, which included several tier-1s and suppliers, suggests that virtualization is still in its early days, and not ready to deliver real business value yet. Consistently, CSPs noted:
- Dedicated NFV-I
- Work-flow based orchestration (not entirely model-driven)
- Partial, but insufficient, policy flexibility (a necessity for context)
- NFVs architected like their physical predecessors, running on “fat” VMs
- HA and scaling built into the NFVs themselves
- Dedicated (proprietary) VNF-Ms
Fundamentally, this limits flexibility and automation. We can’t take advantage of spare capacity easily (reduced capex utilization). We have poor performance per CPU (higher unit capex). We are not agile with fully hands-off, self-realizing, context-aware instantiation of NFVs and Services (lost opex savings). We can’t react to mass-scale short-term needs and changes (agility).
Recently many have commented in print and at conferences that virtualization does not appear to be delivering the promised results. While superficially true, we feel that more important is the understanding that current state of the art is “necessary” – and good progress, but it is far from “sufficient”. Sufficiency demands a set of technical and operational characteristics (see first list above for an incomplete but good summary) that have not yet been widely implemented. The good news is that these should be well understood – we can learn from established and proven practices in other industries, from the hyper scale web providers, to the maturity reached by control theory across myriad industries.
For those seeking more details and background, existing and upcoming ARG research investigates these best practices necessary for success. We also encourage you to contact us if you have supporting or conflicting evidence to share. We’re all in this together!
(links to associated research below)
Free White Papers:
Major Reports, by Subscription (interested in a trial, contact us)