Hybrid Flash Storage Arrays: When and Where?

Hybrid flash storage arrays are more expensive than other options, but in certain scenarios they are clearly the best choice.

Hybrid flash arrays achieve higher performance and lower latency than HDD arrays and cost less than most all-flash arrays. However, they are more costly and add more complexity than traditional HDD arrays. Given the added cost and complexity, what types of applications and environments can most benefit from hybrid arrays?

First of all, let’s look at the problem that flash storage arrays were created to solve: the impact that huge data growth rates are having on the data center. One way to look at the issue is the “3 V’s”: volume, velocity and variety.

Data is growing 100%-plus year over year many date centers. These massive data volumes represent a big challenge to maintaining storage capacity and performance, and managing costs around purchasing, maintaining, data center real estate, and energy costs. Data is also growing at extreme velocity. With 100% growth or more a year, IT struggles to just keep up with growth. Strategic storage decisions are enormously difficult to make in this fast-moving environment.

Now add the tremendous variety of data that IT must store. There are text and digital files; audio and video; mobile data growth; and one of the biggest data producers in history: machine-generated data. For example, a single airline passenger trip generates gigabytes of data on the flight details, the passenger’s identity, security, seat assignment and frequent flyer status, the weather, the airports, and more. Multiply this single passenger’s data by tens of thousands of people flying every day and you begin to see the sheer scope of massive data generation.

Data and IO

Every time data moves between server and storage it generates Input/Output (IO) activity. Fast-growing data volumes, increasing performance velocity and a huge variety of data types all jack up the amount and speed of IO that networks and storage systems are expected to handle.

The problem is accelerated in server virtualization environments. Virtualization is a fine thing for managing server and application growth, but virtualized servers generate even more IO going to a few storage resources. Technology development routinely increases the amount of IO that servers produce, but due to mechanical constraints storage cannot keep the same pace.

There are IO bottlenecks at every stage of the server-to-spindle pipeline including controllers, fabric, and physical servers themselves. But by far the primary bottleneck is storage due to the mechanics of hard disk drives.

The application makes an IO request to the operating system, which directs the IO to the storage system. The storage system determines the optimal placement on disk for the incoming IO and moves the disk drive heads to the location of the incoming IO. Meanwhile the disk platter constantly rotates beneath the heads.

That is a lot of physical activity going on, which severely limits storage performance time. Furthermore, IO writes often break the single IO into multiple IOs in order to fit across non-contiguous locations. Latency due to slow seek speeds affects both writes and the reads.

Yet HDDs are extremely limited in how much they can improve performance speeds. Thanks to their mechanics, they top out at 100-200 IO/s (IO per second) and 3.6 millisecond (ms) seek time. So what happens to the single disk drive to make it produce up to 200 IOPs? It’s striping, clustering, and parallelizing for all it’s worth across multiple disk resources and storage controllers.

Yet even those advanced storage techniques can only go so far. The result is a serious storage bottleneck due to seek time and rotational latency. Meanwhile IO requests in the thousands, tens of thousands and even millions are hitting the storage controllers at any given moment.

Flash to the Rescue?

Flash technology is the most widely accepted solution for redressing the IO bottleneck. Flash covers a lot of territory: some environments deploy PCIe flash cards as server-side caching technology and SSDs may be placed up and down the computing stack at the server, networking or storage levels.

Flash arrays differ greatly in their IO rates and latency. Hybrid arrays combine flash cache and flash tiers with SSD or HDD storage. These systems offer performance speeds up to 100,000 IOPs at 3-5 ms latency while retaining storage capacity in the array. All-flash arrays will usually be considerably faster with throughput speeds well above 100,000 IOPs with sub-millisecond read latencies, but will half about half the capacity or less than a hybrid array.

Prices matter, too. In general the faster the flash system the more expensive it is. Flash prices are dropping and some array makers claim that their all-flash array can be cheaper than a hybrid. (Although all flash is not created equal, and an all-flash array with consumer-grade flash might be slower than a hybrid array.) In any case, IT must always weigh service level requirements against expenses.

Investing in server-side flash and all-flash arrays can be justifiable for critical applications requiring intensive storage processing, but chances are IT will not be able to buy enough of it to store hundreds of TBs of data from applications like databases, OLTP, SharePoint, Exchange, and virtual servers and desktops.

These are critical applications that may outpace HDD array speeds but do not necessarily require pricey all-flash alternatives. This is the sweet spot for hybrid flash arrays. Still, even hybrid flash arrays introduce expense and complexity into the storage environment.

Are they worth it? The answer is: It depends.

· Question #1: Is application performance affected by storage IO bottlenecks?

If the answer is no, then stick with your HDD arrays as long as they work for you. If, however, application performance starts to decline then look to flash storage technology in one or more of its configurations: server-side flash caching, virtualized HBAs, all-flash arrays or hybrid arrays.

· Question #2: Do you have high storage capacity needs in addition to performance needs?

Server-side flash and all-flash arrays best serve business-critical applications with high IO and low latency requirements, whose data stays active over a long period of time, and that does not need to share storage with additional applications. Applications that require higher IO processing than HDDS can give but that generate high volumes of data will strongly benefit from hybrid arrays.

· Question #3: Are your legacy arrays aging and it’s time for a technology refresh?

If you are going to be replacing storage arrays then by all means look to adding flash into your environment. In many cases it is worth the higher cost if you can make the business case for high performance, low latency and high capacity storage.

Hybrid Array Vendors

You will find a bewildering array of vendors. Know what you need before you shop around: many hybrid vendors also offer more expensive all-flash arrays, and sales reps would love to sell you those. Do not let them talk you into all-flash if what you need is low latency, high IOPs performance, and high storage capacity.

IBM, HP and EMC all offer hybrid flash arrays as part of larger storage lineups. IBM’s XIV Storage System offers a hybrid model, and IBM SAN Volume Controller (SVC) centrally manages its flash device line. EMC offers all-flash and hybrid flash choices in its VMAX, VNX/VNXe and Isilon lines. EMC’s key marketing message is that its customers can graduate from HDD to hybrid to all-flash with relative ease; nice if the customer prefers to stay an EMC shop.

HP offers HDD, hybrid or all-flash arrays within the 3PAR product family. NetApp’s FAS3250 is a flash hybrid array, which NetApp positions as the right choice for serving applications that top out at 150,000 IOPs and can tolerate up to 5ms latency.

Oracle's ZFS Storage ZS3 is a high performance hybrid array entry. It combines flash and DRAM for high IOPs performance and lower than 1 ms read latency, but Oracle prices it considerably lower than all-flash arrays with similar performance.

The big storage vendors are by no means the only game in town. Nexsan’s NST5000 offers FLASHtier, which combines a high performance flash cache with DRAM to provide highly efficient flash-based processing for full worksets. American Megatrends 3500i comes as a flash-only or hybrid model combining SSDs and SAS disk.

Tegile Zebi HA2800 offers expansion chassis for SSDs and SAS drives that can expand an all-flash array into a high capacity hybrid. Nimble’s CS line uses multi-level cell (MLC) flash cache with economical backend SATA disk. Tintri offers a hybrid storage array with SATA drives but does it with a twist: its array targets VMware environments by mapping IO requests directly to the VMs.

Some PCIe flash card makers are also getting into the action. NexGen is flash card maker Fusion-io’s hybrid flash server storage. NexGen combines PCIe flash cards with flash memory and SAS disk in a server configuration. Nutanix’ NX family combines PCIe with flash and Intel processors for virtualized environments. The 3000 model adds storage capacity with SATA SSDs and HDDs.

Hybrid flash arrays benefit many environments by balancing performance and capacity needs at a lower cost than most all-flash systems. The sheer number of offerings can be confusing but also offers a variety of options to best suit your data’s performance and capacity needs.

Photo courtesy of Shutterstock.