Hybrid storage array vendors offer simplicity, ease of use

What are your application requirements?

The way that applications use data can vary wildly. This is one reason that a single metric to characterize performance is useless. Apps such as virtual desktop infrastructure vary from lots of sequential I/O when users boot their system to relatively small amounts of random I/O as they open and save files during the day. Databases generally have lots of random I/O, but some databases, such as Oracle, try to optimize reads and writes, saving writes until there are several to the same file, for instance. Video applications that stream a single video file to many different users have lots of sequential I/O.

Storage systems will allow you to optimize different volumes for different types of I/O. You can set up one volume for lots of sequential I/O for video and another for lots of random I/O. Beyond this, you can set up volumes with different levels of redundancy, such as RAID 1, 1+0 or 5, as well as designate whether a volume replicates to another array or to the cloud or how often snapshots are generated for backups.

Assess the cost of support and hybrid storage array features

There can be a large difference between a hybrid storage system's minimum and maximum configuration when looking at supported capacity and overall cost. Scalability is determined largely by the number of controllers in a system. Controllers are devices that manage the disk drives and include interfaces that connect the drives and the hosts. At least two controllers are required in an enterprise hybrid array to ensure availability.

Most vendors offer basic features, such as automated tiering, replication, compression and encryption, in the base price. But you may have to pay extra for more advanced features, such as many-to-one replication.

The cost, however, depends more on how much flash you include than any set of included features. If cost were not a factor, most systems architects would choose the performance advantages of all-flash, but not every organization has the budget for that. There is little point in paying for a storage system that can respond in a microsecond if the network response time is 40 milliseconds. The application should dictate performance, not the other way around. A higher percentage of flash will give you faster response times and lower latency.

It is hard to get a good handle on how much a hybrid flash array will cost without talking to vendors or their resellers. Pricing can depend on factors such as which reseller you purchase from, your discount level based on previous purchases from the same reseller or vendor, or the number of days left in the vendor's fiscal quarter. Vendors that sell through resellers often will not give out even recommended pricing, as the resellers want to be able to charge as much as the market will bear. Getting a firm price often requires an extended dialogue with a sales rep, and if you tell them you have decided on another product, they may suddenly offer to reduce the price. And add-ons can inflate the price dramatically. For example, you may find that a support contract costs 20% of the cost of the array every year, or optional software features, such as replication, are either no cost or are a substantial addition, depending on the vendor.

Evaluating the cost of support and software updates can be difficult. Some vendors offer these for free, rolling the cost into the price of the hardware. Given that operating costs, which include support and updates, are often treated differently from upfront capital costs, this may not be a good thing. Longer agreements can be considerably less expensive. For example, a five-year agreement may be half the per-year cost of a one-year agreement.

Discounts on the upfront costs can also vary widely. For example, some vendors may offer discounts of 30% or more at the end of their fiscal quarter, while others seldom will go much below their suggested retail prices.

Another issue to look at is the sales model -- is it direct or through a reseller? Buying direct rather than from a reseller can provide pricing that is easier to understand because there is no commission to pay, but on-site and presales support may not be as readily available. On the flip side, resellers may make it harder to escalate support beyond what they offer directly as communication to the manufacturer going through the reseller can take longer to resolve problems. Also, resellers may not have the depth of trained engineers.

Look to use an ongoing relationship with the reseller or vendor. Often, a hybrid storage array that is blindingly fast when installed loses speed after a few months of continuous use. Tuning the array for specific applications can also be an arcane art. For instance, what is necessary to optimize the various applications? A few arrays are application-aware and may automatically optimize performance for a given application. But it never hurts to review the suggested settings with the vendor and check their math by visiting online forums. The performance difference between well-tuned storage and the wrong settings may be orders of magnitude.

Many storage vendors now offer converged infrastructure models. These converged infrastructures are pretuned bundles or reference architectures sold in partnership with server, networking and virtualization vendors. Major storage vendors commonly partner with Cisco for servers and switching and VMware for virtualization. Your storage vendor or reseller can recommend the right size and model of hybrid storage array, servers and hypervisor licenses to fit your workload. Converged infrastructure systems include Dell EMC VxBlock, HPE ConvergedSystem, IBM VersaStack and NetApp FlexPod.

Even if you do not have an array from one of these featured vendors, you may already have a relationship with one or more companies. Existing relationships and a higher total volume of purchases may help you get a bigger discount, which can also carry over to the support and update agreements.

The hybrid storage systems examined here are broken into two categories: enterprise arrays and midrange arrays. These distinctions are somewhat arbitrary because some of the midrange arrays can be configured with as much performance, reliability and capacity as the enterprise versions. In some cases, the vendor puts the hybrid array in both categories, with two controllers and a shelf of drives being classified as a midrange array, while the same array with two or four controllers and a dozen shelves of drives is considered an enterprise array.

Examine enterprise hybrid storage arrays

Enterprise arrays typically have at least two controllers, dual paths between the controllers and disks, and redundant power supplies. These arrays generally offer ease of expansion -- not only in capacity but also in performance -- through additional controllers or nodes and multiple paths between nodes.

Scale-up arrays scale by adding capacity to a controller already in use, while scale-out systems add controllers with their own drive shelves. Scale-up may be cheaper and simpler, with no need to add controllers to the management system, while a scale-out system can grow much larger and add throughput and IOPS in addition to capacity. Many systems can do both, adding shelves to an existing controller or more controllers.

Dell EMC

The company's VMAX 100K, 200K and 400K storage arrays can be either all-flash or hybrid. Features include encryption at rest; integration with other arrays, such as XtremIO, CloudArray and legacy systems; and support for VMware Virtual Volumes.

HPE

HPE's 3PAR StoreServ storage features include Fibre Channel (FC) configuration, tiered storage and data protection configured automatically. In addition to the web interface, StoreServ supports a variety of application interfaces and integrates with VMware and Microsoft System Center 2016 to enable configuration through the virtualization management products. The StoreServ system breaks all I/O into small pieces so that smaller requests are not blocked by larger blocks. End-to-end protection of data extends from the host bus adapter to the disk.

With the HPE 3PAR Priority Optimization software, administrators can create service-level agreements, monitor the service levels as needed and set levels for performance, latency and bandwidth. The administrator can use data reduction technologies -- including deduplication, compression and detection of blocks of zeros, or unused blocks -- to increase the effective capacity of the system.

HPE XP7 Storage supports clustering that integrates remote mirroring with a high availability server cluster, providing multisite disaster recovery (DR) and nondisruptive online data migration. Encryption features decrease exposure to data loss, reduce the risk of unauthorized access, prevent unauthorized modification and protect information at the end of its lifecycle by completely deleting data on a specified volume. HPE develops the XP7 in partnership with Hitachi, and the array uses the same underlying hardware as Hitachi Virtual Storage Platform (VSP).

Hitachi Vantara

Tiers of storage on VSP G800 and G1500 enterprise arrays include SSDs, HDDs, flash module drives (FMDs) and cloud. The cloud options include the Hitachi Content Platform private cloud, as well as AWS and Microsoft Azure public clouds. The system includes active flash tiering, which monitors I/O to promote data to a flash tier for lower latency or to move it down to an HDD tier for lower cost. All models run the same versions of Hitachi's Storage Virtualization Operating System, allowing a common set of operating practices to be used across midrange and enterprise models, as well as the all-flash VSP F series.

IBM

The IBM DS8880 series offers self-tuning systems that automatically optimize multiple tiers of storage to deliver maximum performance with a low-cost hybrid system. Tier advisor tools help the data administrator determine the best configuration for their application, including cloud tiers.

Infinidat

The effective capacity of the Infinidat InfiniBox storage array is based on a patented internal data protection scheme known as InfiniRAID and an intelligent compression architecture. Infinidat's compression architecture enables users to store more data on the InfiniBox with no impact on I/O latency. The system is capable of storing multiple petabytes (PB) of effective capacity in a single 42U rack.

NetApp

All NetApp Fabric-Attached Storage (FAS) models run the SANtricity data management software and are built for dedicated, high-bandwidth applications, such as data analytics, video surveillance and disk-based backup. They support multiple tiers, including nonvolatile memory express (NVMe) flash cache, SSDs, HDDs -- including SAS and SATA/Near Line SAS -- and cloud with future support for storage class memory and NVMe drives.

What midrange and entry-level hybrid arrays offer

Midrange may be a misnomer, as many of the arrays covered below can start with a relatively inexpensive and compact system, but scale out and up, with multiple nodes in a cluster and multiple shelves of drives per controller. All of these systems can scale to at least 2 PB and some as high as 16 PB. Many controllers accept multiple interface cards, with anywhere from four to 192 Ethernet ports and the same number of FC ports.

Dell EMC

The Dell EMC Unity line includes the 300, 400, 500 and 600 hybrid models, as well as the 350, 450, 550 and 650 all-flash versions. Maximum drives and capacities range from 150 drives and 2.34 PB for the 300 to 1,000 drives and 9.7 PB for the 600. The system includes an active/active controller architecture and CloudIQ predictive management.

Dell EMC VNX hybrid flash storage configurations include 5% flash for economy; 10% flash for balanced; and 25% flash for performance-optimized. Other services include RecoverPoint and VPLEX replication, which provide replication to both Dell EMC and non-Dell EMC storage. Snapshots can now be stored in the cloud for backup and DR. The storage integrates with VMware, Hyper-V and OpenStack.

The Dell EMC SC Series -- formerly Compellent -- offers three tiers visible to the user, though more than three types of drives can be used. The system automatically puts the drives in the best tier and puts the most active data on the fastest available tier. Virtual drives can be read- or write-intensive or suited to a specific application. Tiering is performed across multiple SSDs and HDDs. Both compression and deduplication can be applied to both HDD and SSD volumes. Snapshot data is also compressed. Upgrades to firmware can automatically upgrade an older HDD system to hybrid and migrate data to the new SSD tier.

HPE

The HPE Nimble Storage Adaptive Flash array combines predictive analytics with flash performance for mainstream workloads. The Adaptive Flash arrays are cited as offering 99.99999% measured availability and enable users to create a multi-cloud environment.

The HPE MSA family of flash arrays, which includes models 1050, 2050 and 2052, are all configurable and support up to three tiers of storage: SSD, enterprise SAS and Midline SAS. The SSD tier can be used in a Read Cache configuration or in Performance Tiering mode, which delivers both read and write performance. Features include virtualized snapshots and remote replication.

Hitachi Vantara

For the VSP G200, G400 and G600 models, tiers of storage include SSDs, HDDs, Hitachi's FMDs and cloud. Storage options include Hitachi Content Platform private cloud, Amazon AWS and Microsoft Azure. The system includes active flash tiering, which monitors I/O to promote data to a flash tier for lower latency or to move it down to an HDD tier for lower cost.

IBM

For Storwize V5000 and V7000 models, upper cache provides fast write response times to the host by being as high up in the I/O stack as possible. Lower cache uses an intelligent algorithm that adapts to the amount of data and provides read caching and prefetching. Transparent cloud tiering on the V7000 enables administrators to move older data to cloud storage to free up capacity on the system. Point-in-time snapshots can be created on the system and then copied and stored in the cloud. An external cloud service provider manages the cloud storage.

Infinidat

The InfiniBox F2000 scales up to 1 PB and 500,000 IOPS and offers all of the software features of the higher-level hybrid flash storage array systems, including an efficient, parallelized data distribution architecture that uses all drives all the time. This architecture delivers high throughput, as well as enables quick recovery from any component failure without impacting performance or reliability.

NetApp

The company's FAS products offer the ability to run both SAN and NAS workloads on unified storage. The FAS2600 hybrid storage array systems use NetApp SANtricity Dynamic Disk Pools (DDPs) to distribute data parity information and spare capacity across a pool of drives, which protects against data loss if drive failure occurs. The DDP rebuild technology uses all the drives in the pool to rebuild a failed drive, which can improve performance during a rebuild.