The ever-increasing size of applications and databases used to run today's enterprises drives the demand for faster systems. In many cases OLTP (online transaction processing), OLAP (online analytical processing), modeling, and heavy-duty video severing have become so mission critical that system performance directly impacts the bottom line.

While this performance challenge has been met by the processor developers, hard drives (HDD) have not kept pace. Performance improvements of CPU and memory have given rise to a "performance gap" between systems and hard disk drive storage.

Even arrays of 15,000 RPM disk drives are at a disadvantage to processor speed because of the mechanical nature of conventional hard drives versus the electronic nature of processor performance. The spinning platters and mechanical assemblies in HDD systems simply cannot present data quickly enough to today's high performance processors (see Figure 1). This latency leaves many commercial applications running inefficiently and users waiting.

All HDD systems rely on a mechanical moving head and platters to access data. When more and more hard drives are arrayed to increase performance, other problems arise for the data center manager, such as power requirements, heat dissipation, rack space, and an ever-decreasing mean time between failures (MTBF).

How well an application performs is generally measured as I/O operations per second (IOPS); and when performance matters, IT managers have many options. They can add server RAM, build bigger hard drive arrays, or optimize their databases. RAM, monolithic RAID, and database optimization solutions work but they are not the best solution for achieving either the fastest performance or the lowest cost per IOPS. The alternative is a technology that's been around for decades - solid state disk.

This less familiar technology is emerging as the front runner for performance, lower cost per IOPS, and reliability in storage. Solid state disk (SSD) has accelerated applications as high as 25x by eliminating the storage performance bottleneck.

Solid state disk systems use fast-access memory chips as their primary storage medium. SSD does not rely on mechanical parts to input or output data in the way that conventional hard disks do. Rather, SSD uses RAM as the primary storage media. Data is stored directly on RAM chips and accessed from them. This generally results in storage speeds far greater than those that are even theoretically possible with conventional, magnetic storage devices. In order to fully utilize this speed, SSDs typically connect to servers or networks through multiple high speed channels such as Fibre Channel.

SSD delivers low latency and high random IOPS compared to HDD RAID systems. Random I/O performance is a more meaningful metric in assessing the application impact of storage performance than the less practical sequential I/O that is typically published.

Unlike conventional memory, SSD systems are built to be non-volatile. Typically, they include battery power and an internal backup disk. In the case of system shutdown or power loss, the battery powers the unit while the data is mirrored from the RAM to the disk. Internal fans keep the unit cool.

Because there are no mechanical parts in the main data chain to the SSD system, MTBF and reliability are higher and maintenance costs typically lower than with conventional storage.

SSD presents itself in an identical manner to disk or RAID, from a software and system standpoint. Hence, no special management or configuration issues arise. In a SAN environment, SSD can co-exist seamlessly with conventional disk and RAID subsystems. Systems with multiple Fibre Channel ports provide additional throughput and support multiserver connectivity via standard switches.

All, or part, of an application's data may be placed on SSD. For instance, database logs and frequently accessed tables may be placed on SSD, while other components are adequately served by conventional storage. Data that resides on SSD may be shared or migrated in the same way as with standard HDD storage. This is because SSD presents itself to the system and OS in the same way. In many instances, the deployment of SSD has led to significant savings from server consolidation and greater storage capacity utilization.

SSD is not a panacea to all performance problems, however. For this reason, customers usually test SSD solutions before buying, and rely on independent third-party benchmarks to prove the vendor's performance claims. Solid state disks currently hold two different records in the Storage Performance Council's SPC-1 benchmark. They have the fastest recorded SPC-1 IOPS performance; and they have, by a large margin, the smallest price:performance ratio. Having said that, they also present much smaller capacities than HDD solutions, which achieve high IOPS performance by incorporating a large number of disks. Therefore, SSD can be not only the fastest, but the cheapest, performance solution when a fraction of total data is slowing down an entire application.

As applications become increasingly demanding and performance is bound by data access limitations, it is becoming a popular addition for savvy IT departments. Growing SAN adoption, falling SSD prices and an increasing performance gap between application performance and conventional storage are all trends that indicate a growing presence for SSD in the data center.