While it has always been the case that IT must respond to increasing business demands, competitive requirements are forcing IT to do so with less. Less investment in new infrastructure and less staff to manage the increasing complexity of many enterprise solutions. And as the pace of business accelerates those demands include the ability to change services… quickly. Unfortunately, older technologies can require months, not minutes to implement non-trivial changes. Given these polarizing forces, the motivation for the Software Defined Data Center (SDDC) where services can be instantiated as needed, changed as workloads require, and retired when the need is gone, is easy to understand.
The vision of the SDDC promises the benefits needed to succeed: flexibility, efficiency, responsiveness, reliability and simplicity of operation… and does so, seemingly paradoxically, with substantial cost savings. The initial steps to the SDDC clearly come from server virtualization which provides many of the desired benefits. The fact that it is already deployed broadly and hosts between half and two-thirds of all server instances simply means that existing data centers have a strong base to build on. Of the three major pillars within the data center, the compute pillar is commonly understood to be furthest along through the benefits of server virtualization.
The key to gaining the lion’s share of the remaining benefits lies in addressing the storage pillar. This is required not only to reap the same advantages through storage virtualization that have become expected in the server world, but also to allow for greater adoption of server virtualization itself. The applications that so far have resisted migration to the hypervisor world have mostly done so because of storage issues. The next major step on the journey to the SDDC has to be to virtualize the entire storage tier and to move the data from isolated, hardware-bound silos where it currently resides into a flexible, modern, software-defined environment.
While the destination is relatively clear, how to move is key as a business cannot exist without its data. There can be no downtime or data loss. Furthermore, just as one doesn’t virtualize every server at once (unless one has the luxury of a green-field deployment and no existing infrastructure and workloads to worry about) one must be cognizant of the need for prioritized migration from the old into the new. And finally, the cost required to move into the virtualized storage world is a major, if not the primary, consideration. Despite the business benefits to be derived, if one cannot leverage one’s existing infrastructure investments, it would be hard to justify a move to virtualized storage. Just to be sure, we believe virtualized storage is a prerequisite for Software Defined Storage, or SDS.
In this Technology Brief we will first look at the promise of the SDDC, then focus on SDS and the path to get there. We then look at IBM SAN Volume Controller (SVC), the granddaddy of storage virtualization. SVC initially came to market as a heterogeneous virtualization solution then was extended to homogeneous storage virtualization, as in the case of IBM Storwize family. It is now destined to play a much more holistic role for IBM as an important piece of the overall Spectrum Storage program.
VMware Virtual Volumes (VVols) is one of the most important technologies that impacts how storage interacts with virtual machines. In April and May 2015, Taneja Group surveyed eleven storage vendors to understand how each was implementing VVols in their storage arrays. This survey consisted of 32 questions that explored what storage array features were exported to vSphere 6, how VMs were provisioned and managed. We were surprised at the level of differences and the variety of methods used to enable VVols. It was also clear from the analysis that underlying limitations of an array will limit what is achievable with VVols. However, it is also important to understand that there are many other aspects of a storage array that matter—the VVol implementation is but one major factor. And VVol implementation is a work in progress and this represents only the first pass.
We categorized these implementations in three levels: Type 1, 2 and 3, with Type 3 delivering the most sophisticated VVol benefits. The definitions of these three types is shown below, as is the summary of findings.
Most storage array vendors participated in our survey but a few chose not to, often due to the fact that they already delivered the most important benefits that VVols deliver, i.e. the ability to provision and manage storage at a VM-level, rather than at a LUN, volume or mount point level. In particular that list included the hyperconverged players, such as Nutanix and SimpliVity but also players like Tintri.
Let's face it: Today’s storage is dumb. Mostly it is a dumping ground for data. As we produce more data we simply buy more storage and fill it up. We don't know who is using what storage at a given point in time, which applications are hogging storage or have gone rogue, what and how much sensitive information is stored, moved or accessed by whom, and so on. Basically, we are blind to whatever is happening inside that storage array. On the other hand, storage should just work, users of storage should see it as an endless invisible resource, while the administrators of storage should be able to unlock the value of data itself through real-time analytical insight, not fighting fires just to keep storage running and provisioned.
Storage systems these days are often quoted in petabytes and will eventually move to exabytes and beyond. Businesses are being crushed under the weight of this data sprawl and a new tsunami of data is coming their way as the Internet of Things fully comes online in the next decade. How are administrators dealing with this ever increasing appetite to store more data? It is time for a radical new approach to building a storage system, one that is aware of the information stored within while dramatically reducing the time administrators spend managing the system.
Welcome to the new era of data aware storage. This could not have come at a better time. Storage growth, as we all know, is out of control. Granted the cost per GB keeps falling at about a 40% per year rate, but we keep growing capacity at about a 60% growth rate. This causes both the cost and capacity to keep increasing every year. While cost increase is certainly an issue, the bigger issue is manageability. And not knowing what we have buried in those mounds of data is a bigger issue. Instead of data being an asset, it is a dead weight that keeps getting heavier. If we didn’t do something about it, we would simply be overwhelmed, if we are not already.
The question we ask is why is it possible to develop data aware storage today when we couldn’t yesterday? The answer is simple: flash technology, virtualization, and the availability of “free” CPU cycles make it possible for us to build storage today that can do a lot of heavy lifting from the inside. While this was possible yesterday, if implemented, it would have slowed down the performance of primary storage to a point where it would be useless. So, in the past, we simply let it store data. But today, we can build in a lot of intelligence without impacting performance or quality of service. We call this new type of storage Data Aware Storage.
When implemented correctly, data aware storage can provide insights that were not possible yesterday. It would reduce risk for non-compliance. It would improve governance. It would automate many of the storage management processes that are manual today. It would provide insights into how well the storage is being utilized. It would identify if a dangerous situation was about to occur, either for compliance or capacity or performance or SLA. You get the point. Storage that is inherently smart and knows: what type of data it has, how it is growing, who is using it, who is abusing it, and so on.
In this profile, we dive deep into a new technology, called Qumulo Core, the industry’s first data-aware scale-out NAS platform. Qumulo Core promises to radically change the scale-out NAS product category by using built-in data awareness to massively scale a distributed file system, while at the same time radically reducing the time to administer a system than can hold billions of files. File systems in the past could not scale to this level because administrative tools would crush under the weight of the system.
Converged infrastructure systems – the integration of compute, networking, and storage - have rapidly become the preferred foundational building block adopted by businesses of all shapes and sizes. The success of these systems has been driven by an insatiable desire to make IT simpler, faster, and more efficient. IT can no longer afford the effort and time to custom build their infrastructure from best of breed D-I-Y components. Purpose built converged infrastructure systems have been optimized for the most common IT workloads like Private Cloud, Big Data, Virtualization, Database and Desktop Virtualization (VDI).
Traditionally these converged infrastructure systems have been built using a three-tier architecture; where compute, networking and storage, integrated at the rack level gave businesses the flexibility to cover the widest range of solution workload requirements while still using well-known infrastructure components. Emerging onto the scene recently has been a more modular approach to convergence using what we term Hyper-Convergence. With hyper-convergence, the three-tier architecture has been collapsed into a single system appliance that is purpose-built for virtualization with hypervisor, compute, and storage with advanced data services all integrated into an x86 industry-standard building block.
In this paper we will examine the ideal solution environments where Hyper-Converged products have flourished. We will then give practical guidance on solution positioning for HP’s latest ConvergedSystem Hyper-Converged product offerings.