NSX-T vs NSX-V... and a Little Bit of Nuage VSP
This week I was very fortunate to be able to participate in one of the first partner sessions VMware have held globally, to learn about and discuss the much vaunted NSX-T1, known as "Transformers" internally. I took a couple of things away from the day at the UK HQ, that I wanted to share in this post to follow on from my last.
There are 3 key points to summarise NSX-T:
- OpenStack Focus
- Different Hypervisors and Compute Platforms (Kubernetes / Docker...)
- vCenter Independence
The popularity and expected continued growth of OpenStack across the enterprise market is the key reason NSX-T exists. In existing VMware environments across business all over the world, VMware are very aware that CIOs and Technology Execs are looking for a way to reduce costs from their businesses balance sheets.
One of the ways businesses are trying to achieve this is by reducing licensing costs, in certain types of environments, and this is one of a number of reasons OpenStack looks attractive. It shouldn't then come as a huge surprise that VMware are looking to try to get a foothold in that market.
Given it's in almost direct competition with their main product set, NSX is potentially almost a perfect complementary product to bridge the networking gaps in an environment that hosts both OpenStack and VMware virtualisation solutions.
NSX-MH (Multi-Hypervisor) was in effect the precursor to NSX-T; so there has been a will to support third party hypervisors for sometime. As mentioned above, it's a good fit product for VMware maintain (and probably increase) relevance given current market trends.
A point to note here is that the only 2 hypervisors supported at this stage, and for the foreseeable future, are ESXi and KVM... though MH had XenServer support this is no longer the case, and not roadmapped - effectively this is now deprecated.
Outside of the traditional hypervisor has shifted significantly towards support for Kubernetes
NSX-V (NSX for vSphere) of course integrates directly into vCenter. Since NSX-T is a product which is built to support multiple platforms, it stands to reason that the NSX Manager is standalone. The upside is that there is no longer a consideration on how many vCenter's there are in the environment.
The defined goal of the NSX-T policy engine is to "Define Policy Once" and simplify and automate delivery of connectivity, security and availability services… to any application, across any platform
VMware see the challenges as Deployment Complexity, Central Management, Visibility and Adapability.
Suffice to say policy is a key aspect of NSX-T, and one that demands much more detail that I'm able to offer in this post.
Nuage was mentioned more than once during the session, and for good reason as it is the most obvious comparable "mainstream" solution in my view.
However it differs in at least one very significant way technically, namely multi-tenancy, which means each naturally targets a different market.
The other clear difference is maturity - NSX-T is not yet released, whereas Nuage have some very high profile customers2 using their solution in anger right now.
The point here is, we're not exactly comparing apples with apples, so in some sense it's unfair, however I think there is value in the consideration.
I think it is fair to say that the target market for NSX, and arguably VMware in general, is firmly enterprise users. In contrast Nuage, with it's telco heritage (Alcatel Lucent and Nokia), is very much suited to the Service Provider arena.
Nuage was built from the ground up with multi-tenancy in mind, whereas VMware pointed out that whilst multi-tenancy "is coming" in NSX-T, it is not currently an feature. I think it is a lower priority purely due to the target market.
Whilst I think it would be a great inclusion, I am cautious on the future of multi-tenancy within NSX-T. The reason I say this is that to me it feels like a fundamental aspect of a product, to be designed from the outset (like Nuage) rather than added later. I do however look forward to being proven to be worrying about nothing!
The Nuage architecture had a fairly simple and consistent picture. I recently drew up a diagram to try and describe it for a customer I was working with:
I don't want to go into too much detail here, but the controller engine is based on (or indeed is) the controller from the ALU Service Routing platform, and there is a consistent picture at the endpoints with VRS and NSG. On top of that, there is good hardware integration for Layer 2 and Layer 3 VTEPs.
With the Nuage solution you feel like you have one solution, and it more or less does everything you need.
NSX-T on the other hand, I found slightly confusing routing picture. As an overview (I'll save detail for another post) There are a number of Tier-0, Tier-1 (each with an SR or DR flavour), as well as edge appliances. Using certain features (like NAT or dynamic routing) determines which appliance you need to use, and has implications in the overall design of the data center.
Now admittedly this was my first deep dive, and the previous caveats about current about it not being released yet of course apply.
These complications aside, I am pleased to say the product is feature rich:
- Tools in the NSX Manager UI - Port Conection, Traceflow, Port Mirroring Session, IPFIX
- Distributed FW
- Distributed Routing (very different from NSX-V)
- Routing to Physical
- Bridging to Physical
- DHCP Server, Relay, Metadata Proxy
One standout point is the use of GENEVE3 rather than VXLAN. It's certainly a fairly bold move, that I am sure will raise some eyebrows.
The elephant in the room is seems to me to be that VMware now have two distinct NSX products to sell and support. NSX-T somewhat treads on the toes of NSX-V as it can be used in a vSphere environment so there is overlap.
NSX-V has been the most successful SDN/NFV solution in the enterprise DC, so if VMware were to want to consolidate, they'll have a large number of customers that would need probably not insignificant migration work.
NSX-V is a key part of the Software Defined Data Center; so having said all the above, it is hard to see it going anywhere any time soon. I'm going to be watching this space very closely over the coming months though; I am really interested to see how NSX-T affects the enterprise market.
Enterprise customers, as I started out by saying, are certainly looking with interest at the OpenStack proposition, and perceived cost savings. Is there a small chance that the product will be so successful, it actually ends up encouraging their own customers to move away from the rest of the VMware stack? Surely not...