Flash storage is getting decidedly more flashy these days. The once exclusive and expensive, high-performance storage technology is staking out an increasingly mainstream storage footprint across the current data environment.
Typically, flash has been the Formula 1 of storage technology, used to meet workloads requiring high-intensity IO with low-latency needs for applications like high-performance computing, database acceleration and data mining support. While it is still used to meet these demands, we find that due to both technology and business model changes, we are able to use flash in a variety of general purpose storage situations, and, in fact, that’s how we are putting it to use in EMC IT.
There are several factors that have made flash a more viable option for a wider array of storage needs:
- Over the last few years, the cost of flash technology has decreased.
- Power and cooling costs for flash are significantly less that for traditional spinning disks.
- Techniques like thin provisioning, compression and deduplication enable users to lower the amount of storage needed making flash more viable.
- Automated storage tiering technology, such as FAST (Fully Automated Storage Tiering), increases the efficiency of using flash as part of a hybrid storage array.
At the same time, IT demands for higher performance storage have increased due to cloud, advances in horizontal applications such as server virtualization and virtual desktop infrastructure (VDI), and the move to IT as a Service to meet increased user demands.
Flash storage options are segmented into three main infrastructure interpretations: in the server as direct-attached storage (DAS) or cache, in the storage subsystem (controller or array) or as a standalone appliance.
The key is to achieve the right trade-off between cost and performance when incorporating flash into your storage mix. For example, EMC IT has found that embedding a little as two to five percent flash into a VMAX or VNX storage system enables a substantial boost in performance from those hybrid arrays. Using FAST, the hot, more in-demand data, goes on flash, while the colder data goes on traditional hard drives.
Another option in using flash is to employ it directly on the server where an application is being run instead of relying on storage in a separate SAN environment. By using PCle media cards, storage is placed as close to the application as possible to achieve maximum performance in demanding applications such as financial trading or scientific computing. Of course, such local flash storage mechanisms have a limited storage capacity.
Then, to reduce the latency of using a full-blown hybrid array or a SAN, All-Flash Arrays (AFAs) have been introduced to accommodate high-performance workloads. Not all flash arrays are created equal though. An AFA platform like EMC’s XtremIO has been fundamentally architected to take advantage of flash speeds, while improving the economics and reliability needed for modern cloud data centers. XtremIO provides rich data services like compression, de-duplication, consistent sub millisecond response times, powerful snapshots, simplified management and deep integrations into hypervisor layer. Typically, AFAs are geared towards simplified data services while hybrid arrays are geared for advanced data services. Going forward, AFAs will incorporate more data services as the capacity and technology capabilities are rapidly evolving. (More on this to come at EMC World 2015.)
While flash continues to grow its footprint in the enterprise data services market, it is still expensive and is not going to provide all the storage needs for a full-blown enterprise. However, it is certainly worth looking at where the application of flash can economically fit into your storage hierarchy.
Flash technology will continue to evolve as storage performance needs change. For instance, IT’s move to the third platform is creating increased demand for in-memory workloads and high performance analytics that will require massively scalable flash mechanisms. EMC is currently working on “rack-scale flash” by applying the technology from its recent acquisition of a company called DSSD in May of 2014.
In a nutshell, flash is a major disruptor in the data center and will restore the parity due to widening gap between the compute and storage speeds experienced over last few years.