This week another vendor, Simplivity, announced that it is going to support KVM and OpenStack… And the list of VSA/hyper-converged vendors supporting KVM/OpenStack is growing very rapidly indeed!
This is an interesting trend and it is worth spending a few words on it. One aspect that most intrigues me is how all the vendors are pushing OpenStack… but it doesn’t make sense to me.
What are we talking about?
KVM is an open source alternative to VMware ESXi and Microsoft Hyper-V. It doesn’t have the same level of efficiency nor the features but it works and all the basic functionalities are available. For some customers (or for specific workloads) it could be a cheaper and good-enough option.
The biggest difference that I see between KVM and ESXi (or Hyper-V up to a certain level) is in the management. In fact, it’s not only about the Hypervisor features but also about how they can be leveraged. The management console (like vCenter), the tool which is used on a daily basis, is a vivid example. And also automation and orchestration suites make a big difference in terms of TCO when it comes to larger environments.
KVM, because of its open source nature, has many different management consoles available but, how many of them can be considered at the enterprise level? As far as I know, none of them come even close to the set of features, scalability, third party support, and robustness of vCenter.
And I can say the same for all other components of the stack that are part of complex virtualization infrastructures. Yes, you can find a good third party tool but, again, the level of functionalities and integration is not always perfect (in part due to the lack of APIs, exposed features or maturity).
This is where OpenStack comes in
KVM alone is not viable in the traditional enterprise, and this is why OpenStack needs to be supported as well, with all its pros and cons. The general (marketing) idea is to push something that, to some extent, is comparable and recognizable as an alternative to VMware for traditional workloads too. I don’t want to argue this point, but the two platforms start from totally different roots positions and we should always take these comparisons with a grain of salt (to say the least).
At the same time, we all know that OpenStack is not seeing a major adoption, also because of its “framework” status. But, on the other hand, it is pushed by all vendors and is continuing its (slow) growing process. And secondly, most of VSA/Hyper-convergence players are now seeing VMWARE (thanks to VSAN and EVO:RAIL) as a direct competitor and are trying to change the game by changing the playground. The problem is that there is a missing part… they are all supporting OpenStack as a separate entity and it’s not going to work!
Hyper-converged or hyper-diverged?
One of the advantages of Hyper-convergence is simplicity. Collapsing various parts of the infrastructure in a single, horizontally scalable, building block is the key point. This kind of approach is particularly appreciated by enterprises of any size while, on the other hand, hyper-scale end users and ISPs have always preferred to build their scale-out infrastructures from scratch.
The basic idea is fundamentally similar (and this is what the vendors are selling you!), but efficiency and optimization have totally different meanings for different types of end users.
On the same wave length of this approach you can find the VMware/Microsoft stacks against OpenStack. The former, thanks to their high level of integration and pre-packaged features, are easier to implement, manage and maintain. The latter gives much more freedom, flexibility and agility at a higher cost in terms of developers, sysadmins, immature features and so on. Again, the difference is between “the product” and “the framework”.
Looking at the overall picture, I can’t see how you can put together a hyper-converged infrastructure with OpenStack. Their approaches clash with one other and are not suitable for the same kind of organization!
OpenStack? Bad!
I’m not saying that supporting OpenStack is wrong per se. But, at the same time, I’m sure we won’t be seeing huge numbers in terms of sales.
For example, startups like Scale computing, specialized in the SMB space, are doing pretty good without it and already offer a good set of functionalities (including backup and DR) without needing an openStack environment on-top for all the provisioning and automation functionalities.
One of the advantages of hyper-converged is also the one step support. You call one number and get all the support you need without worrying about the single component, compatibility issues and so on! What happens when you add OpenStack? Complexity arises, support becomes more difficult and end user experience goes down the drain!
Just think about upgrading OpenStack on your hyper-converged infrastructure. Today it’s Juno, tomorrow might be Kilo, the day after Liberty! It happens every 6 months, and every new release brings new or redesigned modules with it (a nice way to say that it brings immaturity, new bugs and problems alongside maturity, bug fixes and solutions). Can you, as an enterprise, really trust the hyper-converged vendor, which is usually a startup, to support all of this?
OpenStack? Good!
But you can also look at it in another way! By using the Apple approach, and controlling the whole stack, I think the hyper-converged vendor could bring the kind of controlled user experience that enterprises end users want.
In this case I can make the HDS HSP example (I mentioned it a few times already in my latest blog posts but, as far as I know, this is the only one available in this category at the moment). This scale-out appliance is based on OpenStack and KVM. It’s not designed for average workloads, it’s specialized for Big Data, but the idea is really compelling. In practice, you trust your vendor for an end-to-end hyper-converged solution based on open source software. It’s similar to Scale Computing, but with all the advanced features (including APIs!) and the scalability needed for larger enterprises.
In this case, the vendor controls the stack, which also means making it easier to keep tabs on new software releases and giving support is made easier.
Closing the circle
Many hyper-converged/VSA vendors are investing in KVM and OpenStack. I like it, but there are some aspects to consider:
OpenStack is not at a stage where there are well established mature distributions (like RedHat or Ubuntu in the linux world), supporting OpenStack as a third party product can be quite complex and risky… End users are aware of this and, when it comes down to reality, common sense usually prevails. It’s unlikely to be a success in terms of sales.
On the other hand, building a OpenStack-based distribution, integrating it with the rest of the stack, selling and supporting it as a whole, is not an easy task and needs several more resources. Most of the startups in the hyper-converged field are just too small to be credible in supporting such a thing.
In any case I would prefer to see many more of the latter. In fact, I like to believe that we are only in a transitional moment and sooner or later we will have more vendors capable of proposing end-to-end hyper-converged solutions and, above all, end-to-end support!