Network storage arrays have traditionally been comprised of a bunch of hard disk drives. Over the years those drives have become faster, physically smaller, higher capacity, and less expensive. Even so, hard disk drives increasingly fall short in delivering the performance and energy efficiency required in today’s high-demand computing environments.
This chapter takes a look at a technology that offers a higher performance and more efficient alternative to hard disk drive arrays: solid state drives – flash memory-based SSDs. You see how flash storage provides some real advantages and how some vendors create a blend of flash and traditional disk-based storage that address the performance, efficiency, and cost issues in a balanced solution.
Server-based flash has two kinds of solutions. The first is a flash device that’s designed into a PCI Express form factor to fit into a standard expansion slot in a server. The second is a flash based solid-state drive that can fit in to a typical drive slot. Server-based flash can function as a cache in the server or as persistent storage, allowing the server to essentially have extremely fast access to data. Putting high-performance storage directly in the server reduces latency because the server doesn’t have to access data on an external storage network. Server-based flash is high-performance, direct-attached storage.
Server-based flash is often used for frequently accessed data. Some examples of this usage are database indexes or whole databases. In these use cases, data access latency reduction can be as much as 90 percent. Both server vendors and some storage vendors offer server-based flash solutions. While it’s excellent for databases, server-based flash is considerably more expensive than other solutions and has many management challenges, such as high availability and resource sharing across a pool of servers.
Flash memory-based SSDs are significantly faster than hard disk drives and this makes SSDs much better at delivering the performance required for today’s data centers. At the component level, a solid-state drive with a standard Serial Attached SCSI (SAS) interface promises performance many times the promised IOPS of hard disks.
On the other hand, SSDs have a higher cost compared to hard disk drives of the same capacity. Although the cost of SSDs is falling, so is the cost of hard disk drives, which maintains hard disk drives’ status as the best value for large-scale storage capacity.
All-flash arrays are great at delivering very high IOPS at low latency. However, all data does not need to reside on an all-flash array. Cost, performance and capacity are all major considerations in a storage system investment.
Some legacy vendors address the need for more balance between performance and capacity by using data tiering approaches – a means of assigning different types of data to different types of storage media in an array. In a tiered array, you may have a small amount of relatively costly SSD-based storage that handles extremely performance sensitive data, a middle ground tier using high performance disk drives, and a much larger pool of low cost hard disk drive-based storage for less critical data. Depending on the organization’s needs, there could be three or more tiers of storage categories.
Managing tiered storage can quickly become a data management nightmare. The process can be automated to a degree by using software, but the management process continues as an ongoing task.
Storage administrators must understand that data movement across tiers steals storage CPU and data path resources away from handling I/O from production applications. This creates tiering windows, similar to backup windows, that rarely meet the agility needs of the business.
Tiered storage arrays provide different levels of performance for the different categories of data, but sweeping the array for hot/cold data takes time and resources.
If you’re looking for a solution to the cost and performance equation, an answer that offers higher performance, better capacity, and helps keep costs in check is hybrid arrays. Hybrid architectures use fast SSDs and/or dynamic random access memory (DRAM) to cache in-demand data and uses less expensive hard disk drives for low-cost capacity. The result is a faster, high-capacity array at a reasonable price. Using a cache-based design addresses the agility problem data tiering imposes.
Hybrid arrays make sense for organizations that need arrays that balance performance, capacity, and cost. However, on its own the hybrid architecture doesn’t necessarily fulfill all the expectations you may have of a new generation of storage. To meet those expectations, the hybrid array also needs to incorporate a number of advanced features like advanced data protection, compression, deduplication and flexible connectivity.
The way a vendor implements a particular feature can vary widely. Simply saying that a feature exists in a product is no guarantee that you’ll obtain the full set of benefits you might expect.
To more fully understand how different types of storage arrays compare, consider these key points:
✓ Hard disk drive arrays have the lowest cost per GB, but expanding their capacity by simply adding more drives adds complexity without addressing key performance issues.
✓ All flash storage arrays are expensive, but may allow you to consolidate workloads, analyze the business and accelerate your transactions, leading to higher revenues.
✓ Tiered arrays offer a balance between the expense of all flash arrays and hard disk arrays, but can be much more complex to manage and provide inconsistent performance levels to different applications.
✓ Hybrid arrays can provide the best balance between costs and performance, but implementation of important advanced features can vary widely between different vendors.