Welcome again to SDxCentral as we begin 2015 with a bang! If you haven’t had the opportunity yet, I’d encourage you to check out the series of 2014 wrap-up posts by SDxCentral staff:
- 2014: Seven NFV Stories for Beginners
- NFV in 2014: Making the Business Case
- SDN in 2014: Key Stories You Need to Know
- SDN in 2014: Scale, Silicon, and Strong Headwinds
- SDNCentral Contributed Articles: Favorites from 2014
- 2014 In Review: White Box Networking
- SDNCentral’s Top 10 SDN & NFV Stories of 2014
And don’t forget a couple of predictions posts from Craig and Matt as well as Marc Cohn with his ETSI NFV ISG hat on:
- 2015 SDN Predictions: Think SDN Use Cases, Controllers, and White Box
- SDx Trends to Watch in 2015
- NFV Insider’s Series Part IX: Reflecting On NFV Progress In 2014 & Predictions For 2015
An NFV Insider View – an Excellent Crystal Ball
As a true NFV insider, Marc does a great job in his predictions for NFV, covering the following points:
- Orchestration will assume center stage
- Proof-of-concept (PoC) demos will result in trials
- Press releases will yield product releases
- SDN will be questioned no more
- VNF ecosystems will emerge
- OPNFV influence will expand as both a proving ground for open-source projects.
From his front-row seat at the standards bodies, Marc posits that in 2015 we’ll see increased momentum around NFV as a framework and increased deployments of solutions both in-house and from a growing vendor ecosystem at operators world-wide. Based on our own conversations and work with network operators, networking vendors, and other infrastructure vendors, I would agree that Marc’s six predictions are likely to hold true.
NFV 2015 Perspectives from a Different Viewpoint
But I also want to offer some NFV predictions that might not be as straightforward. Some are slight downers on NFV uptake, but I believe these are equally important and likely to impact NFV this year. So here goes:
1. Competing Orchestration Systems Drive Operator Confusion
Without the orchestration systems to provision and manage the VNFs, and also to (just as importantly) meter and monitor VNFs, NFV is not that different from simple compute virtualization — while significant, it’s not necessarily a new trend, since we’ve been virtualizing compute for over a decade now.
Unfortunately, as operators and vendors recognize the importance of the orchestration system, that becomes the new battleground. Multiple alternate offerings will come from different vendors, many of which will be incompatible. Just claiming “my orchestration is based on OpenStack” is not an answer to this mess. We’re already seeing the explosion in flavors of OpenStack distribution from all manner of vendors: Red Hat, Cisco, HP, Juniper, Mirantis/Ericsson, Canonical, … the list goes on. Plus, operators will want to build their own orchestration to meet special needs, to gain a competitive edge, and perhaps to prevent vendor overdependence. We predict lots of flavors in 2015, many more than the market needs, with consolidation not coming for another year or two.
2. Early frameworks on OSS/BSS integration emerge
Once past the PoC stage, operators will look to run early trials within production environments (some have already done so). For larger operators with more extensive existing infrastructure, they will need to figure out the OSS and BSS integration side of the equation.
NFV infrastructure will have to integrate with provisioning systems and billing systems for complete end-to-end solutions that operators can configure and bill for. 2015 will be the year in which best practices start emerging on this front — either cap-and-grow types of approaches that segregate the old and the new, with new services rolling out under more modern NFV management and orchestration (MANO) control, or other integration strategies, including black-box approaches or simply writing more glue code. After all, FCAPS requirements don’t go away simply because NFV shows up, and OSS/BSS remains critical.
3. NFV-washing at both operators and vendors
In some ways, NFV is seeing faster adoption than SDN — no surprise, since a good chunk of the value of NFV is linked to compute virtualization, whereas SDN requires changes to interconnected networks (especially physical underlay approaches). Regardless, we are seeing just about every Layer 3-7 virtual appliance being called a VNF (which I guess isn’t inaccurate), software SDKs for building networked applications becoming VNF components (VNFCs), and everyone else selling to network operators have becoming suppliers of NFV Infrastructure.
We predict 2015 will see acceleration of the application of the “NFV” label at vendors and even at operators — anything that approximates virtualization at the customer/subscriber edge will be an NFV vCPE and anything in the mobile core becomes an NFV vEPC. Certainly, NFV has marketing value and might even raise stock prices. Or as Craig might put it: All your bases are belong to NFV now.
4. NFV price/performance makes operators think twice
Many operators eagerly jumped on the NFV bandwagon because it was a chance to be cool again. What SDN did initially for data center operators, NFV does for network operators. OK, we’ll admit that WAN providers jumped on the SDN bandwagon too — but anyhow, what was not to like about NFV? It promised reduced capex and opex in the long haul, increased flexibility and agility, and fostered innovation for new services. Bandwidth-on-demand gets a new lease on life, virtual services at the edge become possible and cost-effective — and the list goes on.
However, moving from dedicated hardware to an all-software approach usually means less efficient use of the underlying hardware — ASICs in volume are still more cost-effective in moving bits than generic CPUs, even though CPUs are catching up. And early NFV solutions that spawn a virtual machine (VM) for every function that every tenant needs (you want a firewall, here’s a VM; you want an IDS, here’s another VM) will run into cost and performance challenges. By the way, we’re seeing a host of quick “VNF-ication” of existing network applications that have sub-par performance. That’s not going to be fixed quickly in 2015.
Certainly these costs are mitigated by savings from consolidation and reduction in operational costs. And technologies like PCI pass-through and SR-IOV will help. However, price/performance is still a big part of an operator’s ROI consideration. I expect that in 2015, we will see operators think through which functions are best on dedicated appliances (with improved programmability and flexibility), which functions run on bare-metal servers, and which run in VMs.
5. Real-time requirements hamper faster adoption
Network operators are used to providing and meeting tight SLAs (both internal and customer-facing). The move to VM-centric NFV environments without real-time support or resource guarantees in hypervisors will create new challenges.
In dedicated systems, guaranteeing bandwidth and meeting latency caps are much easier to do. Within shared environments, without strong resource allocation or monitoring and dynamic re-allocation support in the hypervisor, meeting SLAs is going to be much more challenging. Today’s virtualization technology worked well in web hosting and for virtualization of enterprise applications where SLAs were less stringent. However, for many NFV applications where timing constraints are much tighter, the NFV infrastructure will have to mature in resource allocation, control, and monitoring. Watch real-time controls in hypervisors and NFV infrastructure emerge in 2015.
6. NFV and clouds converge — it’s just a name
We’re often asked, “What does NFV mean for enterprise data centers?” In many ways, a VNF like a virtual load balancer (ADC) or a virtual firewall has existed and been deployed at enterprise data centers and by cloud service providers. However, the NFV term itself, because of the roots of the effort, is still service-provider-centric.
Certainly, many of the elements in NFV infrastructure more closely mimic those of carriers than enterprise data centers, and enterprise clouds tend to be geared toward hosting end-user applications versus network functions. In particular, NFV infrastructure will be much more focused on network I/O and enterprises on CPU and memory utilization and consumption. However, a virtual firewall can and will be deployed in both, as will a vRouter and just about any other virtual Layer 3-7 appliance. I predict 2015 will see the beginnings of this convergence – Enterprise NFV, anyone?
7. SDx eats everything
We’ve noticed a trend toward cloud architects making networking decisions in the context of application needs, as opposed to network engineers making isolated networking choices. This is true in both NFV and enterprise clouds. Whether a VNF integrates into the hypervisor of choice, whether it supports bare-metal options, what flavor of SDN and virtual networks it support, and what cloud management platform (CMP) it works with might be just as important as the actual VNF feature-set.
What this trends points to is that the underlying infrastructure is the decision driver and that NFV, like SDN, fits within the larger SDx ecosystem. Discussions in 2015 will be less about the VNF feature sets than the surrounding infrastructure.
And that’s our shortlist of 2015 NFV predictions — not all rosy, but we believe they are realistic. We’re big proponents of the NFV movement, but like anything significant, NFV will take time to get there. Growing pains are well worth it. Let’s see what 2015 really brings and if we’re on the money.
Wishing everyone a successful 2015!