As the number of platforms and applications customers have running in their environments increases, so does the complexity of the problems that those customers face while trying to build and maintain a storage environment. In response to this increase in complexity, enterprise storage vendors have worked hard to build out a portfolio of solutions to better address the entire spectrum of customer needs. Although these portfolios are organized in many ways, including storage protocol (FC, iSCSI, FCoE, NFS, SMB) and data type (block, file, object), a primary differentiators is around storage architecture. Deciding on a storage product that scales up or a new architecture that scales out has quickly become one of the most important decision points for customers wanting to build an infrastructure that can keep up with demand. NetApp has focused on providing both options to customers.


When we talk about “scale-up” and “scale-out” storage, it’s important to understand that both designs attempt to deliver the same result to customers. Both want to maintain enterprise-class availability, and both want to provide the performance and capacity that the workloads require. At the end of the day, there can be no compromise on either result. That said, the two architectures take very different approaches to get there. A scale-up environment uses a fixed amount of controller resources and global visibility to all available disks to provide features and performance. A scale-out model uses controllers, or nodes, that have visibility to only the disks they own and are clustered together to create large, dynamic pools of resources.


As customers look to focus on operational and technology lifecycle challenges, scale-out storage architectures have become more popular. While investigating and learning about these products, we hear the same questions repeatedly: What are the benefits of a scale-out architecture? What differentiates the NetApp® offering?


The primary thing that drives customers away from legacy storage architectures is operational complexity. Regardless of the vendor, customers who love their first few dual-controller storage arrays are generally much less excited when they have dozens to manage. Isolating storage capacity to controller pairs that limit the array’s performance and feature set, having to overprovision performance to protect workloads, and the management nightmare of trying to balance performance across many platforms that require a physical data move to resolve workload issues all lead to customers looking for a better way to operationalize storage. It’s these challenges that NetApp is specifically targeting with the SolidFire® AFA, FlexPod® SF, and NetApp HCI platforms, all built on the Element® OS storage platform.


Of course, before we can look at the benefits that a true scale-out design has over its traditional counterparts, we need to be clear about the requirements on which customers must insist when comparing vendors.

Requirement 1: A Real Scale-Out Architecture

The core of any scale-out architecture is the ability to add controller resources at the same time you add storage inventory. Many vendors position a dual-controller, shared disk architecture as both scale-out and scale-up, simply because the underlying file system or data placement algorithm and metadata span multiple controller pairs. Unfortunately, customers soon find that the individual controller pairs still need to be maintained separately. They still need firmware and software updates. They have different lifecycles and refresh rates. Customers find that although they have a large pool of storage, they have gained no improvements in operational efficiency, utilization, or cost.


A true scale-out design such as SolidFire not only adds necessary resources, but also makes those resources available to any volume on the array and clearly defines the amount of resources being added. Knowing exactly how much performance and capacity are being added and being able to direct them toward not just new volumes but existing ones as well gives customers clear visibility into the environment and easy understanding of how and when to scale.

Requirement 2: True Quality-of-Service Controls

Power, without control, is less than useless to customers. After you have an architecture that can deliver millions of IOPS and petabytes of capacity, you must be able to control it. Customers know that workload consolidation is the fastest way to decrease the cost of storage in an enterprise data center. The challenge is that building real quality of service (QoS) is hard. After finding out how hard, most vendors have decided to call their attempts QoS anyway. SolidFire founder Dave Wright put it brilliantly in this blog post: “If corn is a vegetable, why isn’t popcorn? Likewise, if storage performance can be guaranteed, why can’t any storage architecture do it?” If you don’t have consistent, predictable performance (you can’t do math without a denominator), you can’t control that performance. If you can’t control the performance, you haven’t addressed the issue. Unfortunately, most vendors simply bolt features onto the side of a traditional architecture, wallpaper over the deficiencies with marketing and PowerPoint, and say “us too.”


When you have met both requirements, the benefits of scale-out over scale-up become clear and have a huge impact on the day-to-day operation of storage.

Benefit 1: Lower Cost to Scale

In a scale-up architecture, customers are faced with a hard choice when more performance is needed. They can either upgrade the controllers themselves or move to an entirely new platform and incur the cost associated with data migration. A true scale-out design lets customers scale up and down as needed in increments that make sense for the business. There’s never any downtime to add or remove resources, and doing so at smaller increments as needed makes for a much more cost-effective product.

Benefit 2: Independent Scaling of Performance and Capacity

No customer consumes performance and capacity at the same rate or in the same ratio. Having the flexibility to add either resource in the measure needed keeps customers from stranding resources or having to buy resources they don’t need to get the ones they do. Having to pay for capacity to get performance and having to rebuy new controllers just to upgrade the processors in them makes sense for vendors, but rarely is it how the customer wants to consume storage.

Benefit 3: Consistent Performance

In a scale-up system, the amount of controller resources is fixed for the lifetime of the array. As more workloads and capacity are added, those resources are spread more and more thinly, and in most cases the urgency to upgrade is directly tied to performance issues. A true scale-out platform adds controller resources as capacity and performance are added, ensuring that performance is maintained. Of course, the addition of SolidFire guaranteed QoS makes it possible to scale that performance in a linear fashion even as the resource pools grow to significant sizes.

Benefit 4: Data Protection

Despite having the highest percentage of controller resources dedicated to redundancy, scale-up architectures can still have single points of redundancy when it comes to shared disk shelves. In addition, using a traditional RAID data protection scheme can reduce the amount of capacity needed to protect the data. However, this approach requires overbuying disks to use as parity and hot spares, affects performance for an extended amount of time during a disk failure, and requires a large amount of manual intervention to remediate those issues. A real scale-out design can self-heal into the available resources of the cluster without the requirement for redundant components and hands-on maintenance tasks.

Benefit 5: Better Operational Experience

The three- to five-year refresh cycle is both the most important piece of storage vendor lock-in and the worst part of operating traditional storage. At scale, customers can find themselves with a never-ending cycle of upgrades, patches, maintenance, and refresh tasks, none of which improve the ability of their teams to support the business; it’s simply work for the sake of the storage architecture. A scale-out design fundamentally breaks this model by abstracting the part of the storage that is presented to the compute hosts from the resources that sit in the pools being presented. Being able to replace older nodes with newer nodes nondisruptively prevents storage maintenance or capacity management tasks from affecting either application users or applications themselves. It also allows customers to mix and match different node models, letting them adopt the newest flash technology or adjust the ration of performance to capacity any time their environment demands it.


A scale-out storage architecture isn’t a requirement for every storage workload. Smaller environments that don’t need the flexibility and legacy applications that have specialized volume performance requirements continue to benefit from a dual-controller configuration. That said, the balance is shifting rapidly because flash allows for rapid development of the features that customers need. There’s a reason why every new storage architecture that has been released in the last 15 years has been scale-out (3par, Atmos, Cohesity, Equallogic, Isilon, LeftHand, Nutanix, Rubrik, vSAN, XIV, XtremIO, and more). The architecture brings a fundamental operational and cost advantage, and customers are the ones who have benefited.

Get More Information

For additional information on how SolidFire scale-out works, download the datasheet at

Jeramiah Dooley

Jeramiah is a Cloud Architect at SolidFire with a passion for positioning technology in general, and virtualization in particular, as a way to solve customer problems and generate revenue. He is a subject matter expert on service provider business practices, trusted multi-tenancy, VMware vCloud Technologies, and Cisco Unified Communications. When he isn't buried in something geeky, he's home with his wife and two kids trying to see how much trouble they can get into.

Add comment