ARTICLE POSTED September 8th, 2003 Don't overlook solid state disks to boost application performance By Woody Hutsell and Aaron Martz Outside our offices in uptown Houston is one of the city's standard traffic problems — too many traffic lights slowing everybody down. One road over, the problem is too few lanes. Past that, the problem is cars being slowed to a stop so others can access a fried chicken drive-through. Anyone navigating their way to work in intense traffic like we have here and in so many cities is left with the problem of identifying dozens of bottlenecks in a system and predicting the best way to avoid them on a day-to-day basis. Bottlenecks in high-speed computing environments can be equally vexing but much more costly. Computing bottlenecks cost companies billions of dollars each year in lost profits, underutilized infrastructures and time wasted in fruitless tuning exercises. Hard disk: the weak link in the "data chain" Solid state disk manufacturers are saving thousands of companies millions of dollars, but your company probably hasn't tried this technology. If you're like many organizations, you will try everything else: tuning SQL statements or adding servers, processors, memory and spindles, and when all else fails, praying for divine intervention or looking for another job. Why is that? More often than not, the culprit in the "data chain" is the spinning mechanical storage devices that feed everything else. Solid state disks solve that problem by accelerating data to the speed of the servers. A common misconception about solid state disk technology is that it's too expensive. To be fair, we don't use solid state disks at home precisely because it's too expensive. However, our home computers don't need to process every online bet at the world's largest offshore betting site, store every options trade for a stock exchange or analyze huge chunks of the earth with seismic visualization software. What these applications have in common is a thirst for performance. What these companies have in common is a thirst for profits. Drop transactions from the guy trying to place a last-minute bet on the big game, and you lose a customer for life. Delay brokers trying to record options transactions, and you're lucky if all you get is an angry e-mail from the boss. Misjudge a large deposit of oil, and you will probably cost your company millions of dollars. It's not the cost/capacity but the bang/buck that drives these companies to implement solid state storage. 25x performance improvement A single solid state disk can improve application performance by a factor of 25. Real-world users with the same multi-processor servers, the same Fibre Channel fabrics, the same RAID systems and the same software as your data center have seen this kind of performance improvement. On average, customers implementing solid state disks for batch process improvements see a sevenfold performance improvement. Many times, the people who know your application's bottlenecks the best are the application vendors. If you're lucky, you work with one of the growing number of software vendors that are developing, benchmarking or selling their software with solid state disks. Contact your application vendor and ask them what solid state disks can do for your performance. How can you tell whether your application will benefit from solid state disks? One of the best indicators is I/O wait time. When your processor waits on storage, it isn't doing anything else. The application performance benefit of solid state disks comes from eliminating that I/O wait time, thus making optimal use of your processors. In Unix environments, spotting I/O wait time is as easy as running "TOP" or "SAR." These commands will expose the amount of time your processor is waiting on storage. It is best to put these commands into a script so you can monitor I/O wait time over a long period of time, especially peak operating periods. The ability of a solid state disk to improve your performance is totally tied to how much time your processor is currently wasting by waiting on storage. Popular misconceptions aside, Microsoft operating systems provide exceptional I/O performance and can make excellent use of solid state disk systems. It is, however, harder to determine whether I/O wait time is crippling your CPU performance. (Open request to Microsoft: Can you please expose I/O wait time to your customers? For your customers to determine I/O wait time, they have to intuit it by looking at processor utilization and physical disk queues. Isn't there an easier way?) One disk, not hundreds Raw bandwidth needs are the second biggest indicator that solid state disks can help. Video editing, data acquisition, video-on-demand and a variety of other specific applications typically need massive pipelines to keep all their users fed. Users not aware of solid state disks have been forced to buy hundreds of low-capacity, under-utilized hard disks just so they can stripe the necessary bandwidth across them. A single, high-quality solid state disk can handle hundreds, or even thousands, of megabytes per second. Because of the growing disparity between the performance of hard disk drives and processors, the bottleneck in a given enterprise application is often storage. Processors get faster and faster every year, while infrastructure bandwidth pretty much scales to meet the needs of the environments. At the same time, however, the spinning magnetic hard disk has advanced at a relative snail's pace. Rotational speeds have increased, small caches on board have helped, but ultimately the hard disk — a spinning magnetic platter accessed by a mechanical seeking head — can never catch up to the rest of the data chain. Solid state disks work by storing data in SDRAM. Enterprise-capable systems with capacity as high as 64G bytes in a single chassis — supporting one-quarter of a million random I/Os per second and moving data up to 3G byte/sec. — are available on the market. All solid state disk systems include built-in batteries for protecting data in the event of power loss. One major way solid state disk vendors vary is in the mechanism for moving data to the internal hard disk drives. Most vendors only back up data to the hard disk drives when external power fails. This leaves your data at some risk for catastrophic system failures. Other vendors have created mechanisms for constantly moving data to the internal hard disk drives without compromising overall read and write performance. Finally, some vendors only use one hot-swappable internal hard disk drive for backup. What to look for It is best to find a system that uses two if not three hard disk drives for data backup. The persistence of your data in the event of a power outage depends on the health of those disk drives, so it is much better if you have multiple hot-swappable disks for those rare events when external power fails. Solid state disks provide an excellent tool for companies looking to eliminate performance bottlenecks in high-end computing systems. Many companies have delayed looking into solid state disks because of the cost. Now is the time to reconsider your options. Solid state disk prices have dropped 80% in the past three years and now provide more performance improvement than ever, especially in light of the growing disparity between server capability and the lackluster capabilities of the modern hard disk drive. About the authors Woody Hutsell is executive vice president at Texas Memory Systems, a leading solid state disk manufacturer. He can be reached at woody@superSSD.com. Aaron Martz heads up communications for Texas Memory Systems. He can be reached at aaron@superSSD.com. Feedback: feedback@snwonline.com Questions: questions@snwonline.com Technical difficulties: webmaster@snwonline.com