Network disaggregation, unbundling network operating system software from switching hardware, is a growing trend. We’ve seen varying degrees of disaggregation from thought leaders such as Big Switch Networks, as well as full-stack suppliers like Dell and HPE. Even networking specialists like Juniper have disaggregated offerings, with rumors abounding around Arista, Cisco, and others.
Avaya’s networking division has been quiet of late, but it is clear now why. The networking team has been heads down on a disaggregation project. Randy Cross, Senior Director of Fabrics and Infrastructure, chatted with me from China about the announcement, where I suspect it was quite late at night. Thanks, Randy.
From which NOS has Avaya created a disaggregated offering?
Avaya has two families of NOS: the Baystack OS that typically runs on access and edge boxes, and the VSP family that runs on the mainline switching gear. The disaggregated offering is based on VSP.
From the Avaya website, the VSP-based disaggregated offering is called Avaya NOS 4.3, and contains the fabric features based on Shortest Path Bridging (SPB) that we’ve covered in detail on past Packet Pushers podcasts. Randy mentioned that certain of the interesting features in Baystack OS should make it into the disaggregated NOS at some point, although we didn’t get into specifics.
What hardware will the disaggregated Avaya NOS 4.3 run on?
For this initial offering, Avaya has partnered with Delta Networks, an ODM out of Taiwan. Avaya has brought Avaya NOS 4.3 to the entire Broadcom chipset represented by Delta’s switches. Available hardware includes straightforward 48x1GbE switches to full Tomahawk-based 32x100GbE switches with several other choices in between those two extremes.
Avaya will likely partner with other ODMs over time, but chose to focus on one ODM – Delta Networks – for the initial offering to make sure the disaggregated offering works well. However, Randy made the point that porting to other platforms is definitely possible. In fact, Avaya’s disaggregation roadmap already extends to late 2018, with plans for the QA and integration testing required to bring Avaya NOS 4.3 to other ODM hardware.
Incidentally, the amount of testing required for a successful ODM port is an interesting point. I’d have thought a switch with the same chipset is pretty much the same, platform to platform. But according to Randy, the devil is in the details. Yes, they are pretty much the same, but “pretty much” does not mean identical. There tend to be minor differences from ODM to ODM that make the port non-trivial.
What sort of customer does Avaya have in mind with this offering?
Avaya is targeting service providers and systems integrators with this disaggregation offering, which Randy characterized as an “all-in” play. The big idea is to give SPs and SIs a supportable way to offer self-branded networking to their customers. That could be as a switch offering where before they didn’t have one, or as part of a larger offering, where networking is a integrated component.
Self-branding is a key element of this offering. Since Avaya is going after SPs and SIs with their own network offerings, Avaya wanted those SPs and SIs to have their own branding on the gear, shipped to them that way from the ODM. That’s not only the sticker on the switch, but also includes branding of the PROM on the logic board.
SIs and SPs are well and good, but what about enterprises? Enterprises could consume Avaya NOS if they wanted, but Randy says that not that many enterprises are looking into network disaggregation yet. Large enterprises have shown the most interest, but they are slower on the uptake than the service providers and systems integrators who have a business model that maps to network disaggregation more readily.
Notably, ODMs aren’t structured to service smaller enterprise clients anyway. Randy pointed out that ODMs work in large volumes with a smaller number of accounts. Doing business with a large number of accounts buying in small volume just isn’t what ODMs are good at, at least not yet. ODMs will likely get there as the disaggregation market ramps up in the coming years and more enterprises discover the attractive economics of the option.
Supportability is another big driver for customers in Avaya’s crosshairs, which is another reason Avaya chose to limit the ODM relationships to a single one initially. Avaya wanted to be certain that the support experience for disaggregation customers was a premium one.
Avaya has a strong support organization, and disaggregation customers are going to call Avaya first if they have issues. Avaya support will perform the initial review of the customer’s issue, and determine if the problem is software or hardware. Avaya will handle the software issues, but perform a warm handoff to the ODM for hardware problems.
That’s not to say that Avaya waves a cheery good-bye to the customer for the hardware problems. Rather, Avaya will keep the customer’s case open until the ODM resolves the issue, monitoring the ODM’s time to resolution. Avaya is insistent that their ODM partners provide a predictable, solid support experience for their customers, and enforce that expectation by tracking issue resolutions.
How will Avaya NOS 4.3 be licensed?
1. There is a yearly subscription option, most likely attractive to service providers who are trying to lower their costs so that they can compete on price with their switch offering to their customers.
2. There is a perpetual license option, which may make more sense to systems integrators who are including switching as a part of a larger solution they are offering their customers.
3. There will be feature bundle licensing, the details of which are expected to be announced in late June 2016. The intent is to create licensable features such that customers don’t have to pay for features they aren’t going to use. Thus, market-specific bundles are going to be offered. The current tiers and features are listed in the data sheet linked below.
Pick me! Pick me! I want to download Avaya NOS 4.3 to my <insert random ODM switch here>.
Avaya is not offering their disaggregated NOS for general download. Randy says that while they’d like to offer that someday, Avaya isn’t comfortable with this yet. This goes back to the minor differences in ODM switch platforms. Yes, two different ODM platforms might have identical Broadcom switching ASICs, but other minor platform differences exist.
Avaya’s hope is to complete QA testing across a number of ODM platforms, and perhaps offer versions of the NOS tweaked for different switches. Only when Avaya knows what to expect and can therefore set the expectations of downloaders appropriately will the disaggregated NOS be made available for general download.
Is network disaggregation the future?
Looking into the 5-10 year time horizon crystal ball, Avaya thinks that the entire industry is likely to move to a disaggregated model. The driver is going to be a business one. If the cost of obtaining the network part of the IT stack is cheaper in a disaggregated model, then it’s going to win.
The financial models (subscription vs. perpetual license, support costs, etc.) for disaggregation are still evolving. However, once the industry settles down in this area, a cost analysis could look good for disaggregated purchasing. Randy pointed out that the Open Compute Project and related efforts are putting de facto standards in place that make it easier to load a growing number of network operating system options on a wider variety of hardware platforms, all of which helps to drive costs down.
Coupling a cost benefit with the vendor choice offered to customers, and the network disaggregation story should ultimately be compelling to all markets — not just the service providers, systems integrators, cloud builders, and large enterprises of today.