How to optimize PCIe SSD performance

According to Gartner Research, server-side solid state drive (SSD) deployment in the enterprise market is rising by 55% per year and is expected to reach nearly 10 million units annually over the next several years. In applications such as transactional processing, data search, and data mining, the higher performance of SSDs offers tangible, real-world benefits. The cost of NAND flash has been moving steadily downward, and the number of companies servicing this market has been increasing. Because of this, differentiation on value (rather than price) is becoming more difficult, but value is the primary mechanism for storage solution providers to win and sustain their business.

Server and storage systems vendors can add value to what they provide to their customers in the form of performance, scalability, and reliability via their software and hardware architecture. In order for these complex systems to function at their peak, however, two of the largest contributors to these added values—the SSD controller and interconnect—need to be matched and tuned. At the most fundamental level, component vendors need to share a common vision of how a system should be designed and deployed. One of the most important decisions is the interconnect used to get the data from storage to host. PCI Express (PCIe) is rapidly becoming a key interconnect in enterprise SSD storage, and is expected to account for more than one-third of the server SSD market by 2015, according to Gartner.

Let's look at why PCIe-based SSD storage systems (figure 1) are being deployed, and why it is important for interconnect and flash controller suppliers to work together to achieve the best results.

One key factor governing system performance is latency. A common misperception is that data and storage are the same; they are quite different. The only data that is actually being processed by an application at any given time is the active dataset in host memory. All other data is just storage, whether the medium is an SSD or a hard disk drive (HDD). With modern servers possessing powerful multi-core CPUs and including plentiful DRAM, a common bottleneck is the latency of moving data from the persistent storage device to system memory (figure 2). When the processor has to wait to get the data it needs, cycles are wasted and cannot be recovered.