The benefit of SSDs using a disk drive form and connection is that the SSD will work with improved speed without any change in either application or OS software. However, this form factor and connection was designed for disk drives that operate in milliseconds (10-3 seconds) and emasculate SSD latency (potentially measure in microseconds, 10-6 and throughput (thousands of IOPS compared with 100s).
One solution used by companies such as Fusion-io is to use a PCIe connection. This still has functional constraints, such as without the ability to be hot-plugable and without dual-port capability and performance constraints.
There are tremendous advantages in putting a layer of very high-speed persistent storage (persistent storage retains data when the power is removed from the medium, and includes disk drives and Flash storage) between the traditional SAN (very slow) and RAM (which is very expensive, has addressability limitations, is volatile, and can only be made persistent with very expensive battery back-up extensions). This is particularly helpful for workloads that require very low IO latencies and/or very high bandwidth. Examples of this are database and HPC workloads.
A new working group called the "SSD Form Factor Working Group" has been formed. Dell, EMC, Fujitsu, IBM and Intel are the "key promoters" of a working group to improve the PCIe form factor and connectivity. The charter is to promote enterprise storage usage of PCIe SSDs, by enabling serviceability, high-availability, ease of integration, interoperability and scalability of solid-state storage. The key Focus Areas are form factors, connectors, and hot-plug behavior. This aims to:-
- Allow multiple power sources for improved availability and recoverability
- Allow external access of PCIe cards and allow hot-plugability
- Allow dual ports for improved availability and recoverability
- Allow the ports multi-protocol access to the card using PCIe, SAS and SATA protocols, which will improve availability, flexibility and recoverability
- Provide a management bus which will allow management across storage close to the server, storage on the SAN, and cloud storage
- Improve latencies in the IO protocol stack and NVM to improve performance and throughput of current and future flash (or even PCM) storage technologies.
Fusion-io, STEC, QLogic, Emulex and others are "contributers" to the working group (Figure 1 shows the promoters and contributers to the working group). Vendors such as Fusion-io were the original innovators of flash and PCIe and specifically Fusion-io is growing very rapidly and has bolted out to an early lead especially in emerging Web 2.0 markets. EMC has an interest in slowing Fusion-io's momentum and improving the functionality of its storage arrays by including higher function read/write flash cache. Oracle and HP are key players that are conspicuously absent and should be encouraged to participate, as improving the functionality of PCIe flash will raise all boats by speeding the adoption of system technologies that provide instant-on and instant recovery. This in turn will drive ease of use through the roof by allowing much higher IO rates, in the same way as Apple has done with flash on the iPad. Ultimately such a working group will be all-inclusive or destined to fail.
Action Item: This improvement to the PCIe specifications will be a significant improvement to storage access and management in future systems. The enterprise storage topology at the array and SAN level will morph into a distributed storage topology, with storage within the server, on the SAN and in the cloud. CTOs should be asking questions of their Infrastructure 2.0 vendors about their commitment to these standards, and their commitment to storage management systems that will allow seamless management of data between server, SAN and cloud.
Footnotes: More information can be found at the SSD Form Factor Working Group Web Site