Private cloud storage 101: Key components and hardware options

Cloud definition

As already mentioned, private cloud storage echoes the offerings we see in the public cloud. The classic definition of cloud was set by the National Institute of Standards and Technology (Nist) many years ago, and we can use this definition as a framework for private cloud storage.

Self-service: Users (in this case, the business) should be able to request storage resources and have them provisioned with as little impact as possible. A self-service storage architecture needs automation in the form of application programmin interfaces (APIs) and command line interfaces (CLIs), which is integrated into the application provisioning process.

Elasticity: Users shouldn’t be concerned with scaling storage resources up or down. In effect, private cloud storage should appear infinite and be available on demand. In the public world, this requirement is pretty clear, because public cloud service providers (CSPs) continue to see continuous growth in volume of new customers (for now at least). In private cloud, budget constraints and business growth means IT has to put much more effort into being efficient with storage resources, while providing the illusion of elasticity.

Resource pooling: Rather than islands of storage, deployments should be pooling resources and getting the benefits of economies of scale. This represents a challenge to implement, because hardware isolation offers better security and performance management. As a result, features like quality of service (QoS) and Multi-tenancy become important to deliver private cloud storage.

Billing: Typically, this is seen within public cloud as delivering a measured service, or charging for usage. Charging for usage can be an issue for many IT organisations that have evolved to implement IT resources using project-based deployments.

Performance: Although not strictly a Nist definition, performance is more significant in private cloud than public cloud. Applications in private deployments tend to be monolithic and latency sensitive, which makes input/output (I/O) performance all the more important.

Service levels: For service-based deployments, we need to have service levels, if not service-level agreements (SLAs) in place. Service levels form the basis of a service catalogue – a list of offerings and their descriptions.

Private cloud transformation

Moving to a cloud-based deployment for storage has many advantages.

Users can be abstracted away from needing specific hardware, allowing IT to optimise the products they buy. IT can choose the most efficient refresh-and-replace cycles, choosing whether to replace hardware or extend maintenance. This is done without affecting the user financial model as charging is typically done on a consumption-per-month basis.

The on-demand nature of private cloud means businesses can be more agile, because the “human factor” can be removed from deploying storage for new and existing applications. This is where API-based management becomes critical.

But moving to a private cloud model is not without its challenges. Business processes for application deployment may be based around projects and not structured to handle ongoing payment for resource usage. This may mean significant changes to the process of budgeting for IT across the business.

Building a cost model for IT also changes. Where previously the business may have been charged simply for the cost of infrastructure, now a more complex model needs to be developed that considers all the costs – capital expenditure (capex) and operational expenditure (opex) – and translates that into a recurring (opex) monthly charge. If the business doesn’t need as much storage from one month to another, how will this shortfall be made up? This is an issue for many IT organisations, and one that is a significant factor in those choosing not to adopt a private cloud model.

Private cloud building blocks

In putting together a strategy for private cloud storage, the main considerations are a service catalogue, multi-tenancy and quality of service (QoS).

Service catalogue: A service catalogue puts in place generic definitions of the requirements of the user, without being hardware-focused. The catalogue references service levels, such as tiering, performance and availability, without discussing specific products.

Multi-tenancy: Storage will be shared to get economies of scale. Multi-tenancy features mitigate risks such as “noisy neighbour” and segregate applications from a security perspective.

QoS: Quality of service is important to ensure users get the resources they pay for. The ability to dynamically change QoS settings is a real benefit for private cloud and a differentiator over public cloud, where customers are generally expected to repurpose or reprovision their application to increase I/O performance.

APIs: This has already been stated, but needs to be re-emphasised to highlight the quality and specific implementation of APIs or CLIs. Some storage suppliers have added APIs as an afterthought, making it difficult to manage storage from multiple interfaces (for example, GUI, CLI and API together). APIs should be native and inherent to the platform.

Analytics and monitoring: Inevitably, performance and configuration problems can occur. Most IT organisations will have tools already in place, but these may need revisiting with a private cloud perspective. This can mean building out better reporting/billing, establishing thresholds with automated responses and a long-term capacity planning strategy.

Process: The term “process” seems very generic, but there are a lot of traditional tasks that would affect storage that are not seen in the public cloud. This includes, for example, maintenance and downtime for patching and upgrades and hardware replacement. Ensuring 100% uptime with minimal or no maintenance means implementing processes for live workload migration, including working with server and virtualisation teams to build joined-up solutions.

Private cloud products

HCI: Hyper-converged infrastructure integrates the functions of storage into the virtualisation platform, providing one scale-out deployment model for compute and storage. Typically, the HCI storage component is highly integrated, with little or no work needed by a storage administrator to manage the solution.

VMware provides HCI in the form of Virtual SAN, which can be enabled through a software licence on each vSphere host. The customer is responsible for managing the hardware itself and any associated internal server storage. Dell EMC solutions use Virtual SAN to build HCI appliances that come in a range of configurations.

HPE offers HCI in the form of SimpliVity appliances. Storage is integrated tightly into the architecture, with dedicated hardware that implements performance and storage capacity optimisation features. HPE also offers solutions built around the Synergy architecture.

Nutanix offers HCI solutions that implement a scale-out storage layer across a multi-node configuration. Storage offerings include block-based devices integrated into the hypervisor, external block-based storage and file services.

NetApp recently introduced an HCI range of products, based on the SolidFire storage technology. Solutions scale both storage and compute independently, which fits an architecture between HCI and converged offerings. The SolidFire platform offers native features such as QoS and API-based management.

Convergence: These are solutions that integrate storage into a hardware “stack”. Typically, converged offerings provide good automation features, but are weaker on scale-out, due to the pre-configured nature of converged infrastructure products.

Hitachi Vantara offers a range of converged solutions based on Unified Compute Platform (UCP) and Hitachi Storage. These products are strong in their software components for private cloud, including Hitachi Cloud Automation Suite and Automation Director. Hitachi also offers hyper-converged solutions based on VMware Virtual SAN.

Appliances

Storage appliances provide another way to implement private cloud storage. Examples of suppliers in this area include Tintri, HPE and Pure Storage.

Tintri offers a range of appliances designed to work with virtual machines. Features such as QoS, replication and data protection are implemented at the virtual machine (VM) level. Tintri management solutions provide automation via API and CLI, as well as federation, making it easy to balance resources across multiple appliances.

HPE 3PAR has many features that implement private cloud, including QoS, API management and storage capacity-optimised features. HPE has extended the ease of management for 3PAR to the storage fabric, with simple device discovery and provisioning for Fibre Channel storage networking.

Pure Storage offers FlashArray appliances with native automation via APIs and CLI. Maintenance and management of multiple arrays can be achieved with 100% availability using ActiveCluster. Financially, Pure offers customers the ability to upgrade systems without the cost and disruption of traditional “forklift upgrades”.

Software-defined

Finally there are software-defined offerings entering the market that provide features of scale-out and automation. SDS suppliers such as Portworx, Hedvig, StorMagic and Starwind offer storage that can be deployed on-demand and in an automated fashion to build scale-out solutions.

We’ve only presented a snapshot of the storage solutions on the market today. Most suppliers are adding cloud-based functionality to their platforms as they recognise the benefits of enabling customers to implement a much more service-based consumption model.