Solid State Storage

Disk subsystems often are the slowest components in workstations and servers. This performance disparity has become even more apparent as processor and memory speeds have escalated quickly, whereas disk subsystem speeds have increased at a much slower pace. For I/O-intensive applications (e.g., video editing, Web services, messaging systems, database management), relatively slow disk storage severely impacts overall application performance and system scalability. For example, if your application uses only 25 percent of your 8-way server's CPU capacity but I/O requests queue up, you aren't using your expensive server effectively.

Although improvements in magnetic-disk technology have resulted in dramatic capacity increases, I/O performance has improved at a much slower rate. Caching I/O controllers, RAID arrays, and caching algorithms in Windows and database management systems (DBMSs) improve data-access speeds. Because caches hold much less data than disk subsystems do, the caching algorithms attempt to predict which data blocks the application needs and caches only those blocks. In an effective cache, the correct blocks are present most of the time. But, depending on the locality of the application's read requests, the blocks might reside in the cache for only a few seconds before they're replaced. If the blocks are bumped from the cache, the next time the application requests them, the application must wait to retrieve the blocks from the hard disk or array. And if the requested blocks aren't sequentially located on the disks, the performance hit can be substantial.

Obviously, you can tune application performance to minimize system I/O and improve overall performance. But tuning application performance can be a time-consuming process and might not result in the performance gains you're looking for.

Solid state disk (SSD) products improve overall application performance by permanently storing the most frequently accessed data blocks (typically entire files or groups of files) on the solid state storage device. Systems administrators can identify which files are responsible for the highest percentage of I/O requests, then put those files on the SSD. Access times for the blocks that make up those files then can be hundreds of times faster than similar requests from rotating media. And access times for randomly requested blocks stored on the SSD are just as fast as access times for sequential-block requests.

High SSD prices have limited their use to military and other high-priority applications. However, the recent decline in memory costs has sparked renewed interest in SSD products for mission-critical commercial applications.

On a per-gigabyte basis, SSD prices have fallen dramatically but are still much higher than rotating-disk prices. SSD capacities range from 512MB to 399GB; the least-expensive configurations sell for about $1000, and high-end systems sell for $400,000 or more. An SSD large enough to store an entire application and data files could be prohibitively expensive, so systems administrators typically use SSDs to read and write to frequently accessed files (sometimes referred to as hot files). Hot files are typically 1 to 5 percent of a database application.

Before you decide to purchase an SSD, you must determine whether your performance problems are I/O-related. Windows Performance Monitor can help with that assessment. In addition, some SSD vendors offer tools that provide a good overview of system performance. If I/O is indeed creating a performance bottleneck, check your CPU utilization. You can use an SSD to boost I/O performance (which will result in overall application-performance improvement), but only if your CPUs can accept the additional workload. Depending on your application, SSDs can improve overall performance 2 to 10 times.

Entry- and Mid-Level SSD Products
Vendors most commonly sell SSD products in three form factors: PCI expansion cards, directly attachable units the size and shape of 2.5" or 3.5" disk drives, or standalone 19" rack-mount products that you can connect directly to one or more servers or use in a Storage Area Network (SAN) environment. For the largest enterprise applications, vendors offer several capacities: PCI expansion cards typically hold up to 8GB of data, the directly attached drive-shaped units I looked at provide up to 77GB of storage, and rack-mounted products support hundreds of gigabytes of storage. Windows sees these products as standard hard disks. I can't cover all the SSD products that more than 15 vendors supply, so I discuss a few products targeted to the Windows audience.

Most SSD products I looked at use SDRAM and therefore require backup power to maintain their contents during commercial power interruptions or, for SSDs installed inside a server or workstation, during system shutdowns. Data-protection features, as well as the number and type of built-in I/O interfaces, are major differentiators that make some products better suited for particular applications and contribute heavily to the price. The interface technology matters even more with SSD products than with rotating disks—because SSD products provide so much bandwidth, in high-volume applications the interface could become a bottleneck. But regardless of these additional features, SSD prices vary widely, so consider several vendor offerings before you make a decision.