Whether they admit it or not, the emergence of public cloud providers has dramatically altered the playing field for hardware vendors of every type.
Amazon Web Services (AWS) and its competitors opened Pandora’s box by introducing the world to a completely programmatic, scalable, evolving, and pay-as-you-go way to procure and utilise network, compute and storage resources on a global scale.
They have disrupted many layers of the technology industry, from the applications being written to the way companies interact with the infrastructure being used to support those applications.
Nowhere is this disruption easier to see than in the virtualisation ecosystem. For the better part of the last decade, hypervisor companies like VMware, Citrix, Microsoft, Red Hat and others worked hand-in-hand with hardware manufacturers like Cisco, NetApp, EMC, HP and Dell to define both the infrastructure foundation, as well as the virtualised abstraction layer that sat underneath the entirety of the client-server era.
These companies provided a direct link between the enterprise applications, the hypervisor and the hardware. They owned the traditional data centre construct.
It’s that construct, since rebranded as “private cloud”, that is directly under attack by public cloud providers. This will be the battlefield for the heart and soul of enterprise IT for the next decade.
The response to the public cloud threat has been varied, and often reflects the ability for traditional companies to pivot and meet the challenge. Interestingly, erstwhile competitors Microsoft and VMware reacted similarly. This is because they were both uniquely positioned to create a software-defined solution to the problem.
For both companies, the response started with existing enterprise workloads. One of the largest challenges of the AWS public cloud is the fact that getting workloads, and especially data, into and out of an enterprise environment can be both technically challenging and expensive.
Most workloads running on an enterprise-virtualised platform today can’t be easily ported into AWS, and this increases the cost and risk of any migration. As companies with extensive and hard-won experience running mission-critical enterprise workloads, Microsoft and VMware came to much the same conclusion: build a public cloud using their existing platform and allow customers and developers to leverage all of the investment they’ve made in their own data centres as they selectively move workloads outside of their own data centres.
Thus, Microsoft Azure and VMware vCHS were born. Both are clouds that customers can move workloads to without the need to rewrite or re-architect them. They can also be licensed using existing agreements and can be managed by existing staff and tools.
Unfortunately, the traditional data centre infrastructure is now the weak link in this new software-defined world. In each of the public clouds referenced, the focus has been on the abstraction layer and how it interacts with the end users. What’s missing is how the abstraction layer and the applications and tools that sit on top of it interact with the infrastructure directly.
There have been attempts at hardware-based offloading, especially with regards to storage. VAAI is a good example of VMware trying to create a way to let enterprise storage arrays handle the tasks they are good at without requiring the direct involvement of the hypervisor.
But even there, it’s a rudimentary exchange at best: the hypervisor asks “can you do this task instead of me?” and the array responds. If the answer is yes, the hypervisor waits for the task to complete, if the answer is no, the hypervisor does the task itself. This relationship isn’t dynamic, and is ignorant of the reason for and context behind the task in the first place.
So in summary, there is an outside force, AWS and public cloud, being the primary catalyst driving change into the enterprise, yet very little of that change is happening below the cloud management or hypervisor layer. Why is that? Why is it important that the infrastructure layer become more of an asset to the rest of the stack? What would that look like?
The question of why is actually pretty simple: it’s really, really hard to take legacy hardware architecture and retrofit it into something agile and programmatic. In some cases, it’s just a new concept that requires a hardware refresh (like Cisco UCS and its take on XML-defined BIOS policies), but in many cases, especially around storage, it requires a complete reimagining of the platform. It’s no coincidence that most of the innovation in this agile infrastructure space is being done by startups that have no legacy customers, technical debt or margins to deal with.
Why is it important? Because while the best hardware is boring hardware, it’s still a critical part to providing a flexible, reliable and high-performance foundation to handle applications that matter to enterprises. There are times where the best way to handle the demands of an application, or more importantly multiple applications at once, is in hardware.
This is true at the network layer, where the manipulation of packets benefits from proximity to processing resources, the compute layer, where apps can benefit from having specialised GPU resources to handle unique requirements, and most especially at the storage layer.
Storage services can have the most dramatic impact on workload performance, yet are often implemented in such a way that they have no direct relationship with those workloads.
Services like compression, deduplication and quality-of-service are usually “on or off” features when it comes to storage arrays. Best case, a storage administrator will create a volume or LUN, choose the features that need to be enabled, and then a virtualisation admin will map that volume to a data store.
Perhaps the virtualisation team will create manual storage profiles that define the features offered by that data store, but placing and migrating VMs remains a manual process, and they will not have the ability to map application policy equally across the hypervisor and hardware layers. (Of course, it’s not impossible to create programmatic, hypervisor-aware infrastructure, but it is pretty hard.)
Enterprises have come to expect some fundamental features from the public cloud space: simple architecture, linear scaling, API availability and granular application of services.
These features allow an infrastructure to respond to the increased requirements of a workload natively, without the overhead of a bolt-on orchestration engine. They provide the ability for the hypervisor to be both a northbound and southbound policy enforcer.
They enable the next-generation data centre, one in which the hardware, the hypervisor and the application itself all play an integrated, coordinated role in providing the performance and availability demanded by the enterprise.
No matter where a business’s workloads run, the rise of public cloud has ushered in an era of computing defined by a seamless, programmatic experience.
The old, monolithic infrastructure of yesterday’s client-server wave is giving way to a more agile, more responsive, more services-rich and more scalable cloud-based model.
The battle for the enterprise soul is beginning, and inside or outside the firewall, the clouds that can best adapt to the demands of the workloads they are supporting will be best positioned for success.
Sourced from Jeramiah Dooley, cloud architect, SolidFire