2011 Prediction: Aprius
The Emergence Of Solid State Storage In The Data Center
Good news has arrived for solid-state storage in the data center. The commoditization of flash solid state storage will assist enterprises to achieve higher performance for hot data. The marrying of technologies such as Converged Enhanced Ethernet (CEE), PCIe Flash and I/O virtualization will enable flash-based arrays and appliances to deliver high performance without sacrificing economics, scalability and usability.
Most recently, we’ve seen flash storage shift from exotic and expensive to usable performance for attainable prices, expanding the enterprise use cases beyond government and military to database and file system caching and soon the first tier of persistent storage. However, flash storage presents a fundamental change to how storage performs and behaves, and therefore affects how those systems should be architected. Data center systems innovators are leveraging the benefits seen in low latency, fast response times and an order of magnitude increase in I/O operations per second (IOPS) and throughput. They are working around some of the associated flaws of flash storage such as cost, capacity, endurance and scalability, and are rethinking the old paradigms of “direct attached storage” and ”storage area networks.”
In the past, fundamental technology -- single-level cell (SLC) and multi-level cell (MLC) -- within SSD has affected its adoption. SLC is generally known to be fast and durable; MLC is cheaper but less durable, meaning that companies were choosing between reaching high performance and breaking the bank. Now, thanks to advances from companies like Intel and others, MLC is delivering excellent performance within enterprise-class systems at cheaper prices.
Traditional storage protocols, such as SATA and SAS, are being challenged by new approaches. Rather than building on technologies designed for spinning disk and doubling performance, innovators are seeing the opportunity flash presents to triple or quadruple performance. The high performance of flash storage, particularly for caching and random-access applications is often best realized through the placement of the flash storage directly on the host system I/O bus via PCIe-style flash storage cards that provide minimal latency and maximum bandwidth between CPU and storage, without protocol conversions. This technology is truly disruptive and is causing ripples up and down the ecosystem.
In fact we have already begun to think about how this will affect storage systems and the servers that they connect to. We’ve all been part of or witnessed the debate: which storage technology is best for your architecture? The answer usually lies in DAS, SAN or NAS. While the age old debate rages on and architects shoehorn flash storage into current storage arrays in an effort to find the best place to take advantage of the performance of flash storage, we see innovators disregarding the traditional boundaries and proposing hybrids of DAS and shared storage, NAS and SAN, tiering and caching.
Making SSD available in traditional drive form factors and as shared storage in conventional SAN or NAS attached arrays is today’s answer to this debate, but full flash throughput is limited by increased latency across multiple hops in the fabric data path, SAN protocol processing and CPU processing, of both data and control path traffic. In 2011, we will argue, why not make flash storage available as logically ‘local’ devices but physically pooled and shared for scalability, efficiency, concurrent access and high availability.
