Do you find that your storage isn't fast enough?
Are you buying lots of small, high-rpm drives for time-critical transactional databases?
And how would you react if someone told you that a technology already exists that provides 30 times more input/output operations per second (IOPS) than 15,000-rpm Fibre Channel drives?
After years of concerns about speed and reliability, this technology – the solid state drive (SSD) -- is finally eliminating the slowest component in transactional databases: the disk drive.
There is one glitch: SSDs are about 50 times more expensive per gigabyte than a 15,000-rpm drive.
Perhaps you cannot contemplate paying that price for performance -- but speed is worth big money for many companies. Many brokerage houses, for example, pay huge penalties on trades that aren't processed within a specific time. At other companies, faster-responding business analytics engines could enable near-real-time feedback for advertising or product placement changes, driving multimillion-dollar, time-sensitive business decisions.
So if shaving milliseconds off critical transactions can make you money, SSDs may well be for you.
Solid-state drives deliver speedy response
It is worth noting that solid-state storage isn't new. Disks and arrays have built in memory caches to buffer the performance of slow disk drives for decades. The problem was that memory was much more expensive than spinning platters, so caches stayed small and were of limited utility.
Even when memory prices fell, nonvolatile memory just wasn't fast enough to rival disk. To protect the performance and data integrity of the data, solid-state storage was often just fast volatile memory front-ending traditional disk drives.
Recent developments have erased those problems and solid-state storage I/O performance (IOPS) is now linear, meaning that the SSD disk response times remain consistent all the way to their maximum I/O capability. In comparison, traditional disk drive response times get progressively worse as IOPS increases.
Some niche players, such as Texas Memory Systems, have since developed 100% solid-state storage arrays. But they require you to put 100% of an application's data on SSD or spread it across SSD and traditional arrays to tier within a database. There weren't many good hybrid solutions.
Recently, though, EMC changed the dynamics by announcing SSDs for its DMX4 high-end storage array. Now a single box can provide a tiered storage architecture, including Tier 0 SSD to Tier 1 and 2 Fibre Channel and Tier 3 SATA drives. Others will surely follow because the value proposition is simple: A single database may have many regions of data, each requiring different performance characteristics. Indexes may benefit from the supreme performance capabilities of Tier 0 SSD, and the same database may have archive regions living on Tier 3 1 TB SATA drives. One application, four tiers of storage, all in one box.
Good candidates for SSD
Despite its obvious performance appeal, SSDs are still a lot more expensive than traditional storage and they're not a good fit for all data.
If you want them to make a difference, you need to understand how and when to use them.
Applications like financial trade clearing systems, where speed is at a premium, are an obvious application. Situations where you stripe lots of small disks and use only a small amount of the storage on each drive to guarantee a certain access density offer a chance to use SSDs to reclaim those striped and underused disks for normal storage.
Other critical, time-sensitive applications that justify SSD will probably be transactional databases with small and/or random read/write profiles. High-transaction databases usually have small (4 KB to 8 KB) records that are accessed randomly, making disk array cache pre-fetching and read-ahead algorithms ineffective. SSD, on the other hand, is well-suited to such access rates. These databases may also frequently log read access for replication or rollback transactions -- both of which are random reads and could benefit from SSD. Finally, business-critical transactional databases don't get a lot of downtime and may be highly fragmented. Fragmented databases produce random I/O, causing disk drives fits while solid-state drives soar.
It's important to note, though, that solid-state storage isn't always about performance. Sometimes it makes sense in niche applications that require ultra-high reliability. For example, SSDs are already common in aerospace applications, high-vibration environments and desolate locations like Antarctica. This is because solid-state drives have no moving parts, which usually translates to a higher mean time between failures (MTBF). These high levels of reliability may be necessary in remote or inaccessible locations, as well as under extreme temperatures, pressures and g-loads. SSD is also a good fit in mobile applications where normal disk storage just won't work.
Even with the higher cost per gigabyte of solid-state storage, customers may be able to save money by improving performance or reclaiming underutilised performance disks and replacing them with solid-state drives. The best plan is to present the performance and pricing data to your customers and let them decide which technology best suits their needs.