Recently, I was invited to speak at the Gartner Symposium in Orlando, Florida. The presentation I chose to deliver was on the importance of flash in mode 2 data center designs. (You can view the recording of that presentation here.) What’s “mode 2”, you say? In a nutshell, it’s a Gartner classification of approaches to data center infrastructures. Mode 2 is centered on converged/hyper-converted compute and storage resources that run web-scale applications, lending themselves to recent techniques and methods such as SAP HANA, NoSQL, Cassandra, Spark, MongoDB, Hadoop, and other workloads, including many distributed file and object systems. Mode 1 is the traditional approach that uses discrete servers, a data fabric (e.g. fibre channel or SAS), and SAN storage—the so-called ‘scale-up’ approach. Gartner advocates for enterprises to run both modes in parallel to begin solving efficiency problems and move into a scale-out, node-based infrastructure. Cutting over from one mode to the other is problematic for many enterprises, hence the parallel approach.
My presentation specifically describes how flash is the critical technology for use in mode 2 data centers. Flash is not just an ‘enabling technology’—a phrase we’ve heard quite often—it is a technology that allows such workloads to scale and thrive, not just run. We’ve been stuck in mode 1 approaches for the better part of two decades now, and while those served us well running relational databases, email, and other classic back-office applications, it is a futile exercise to use them for workloads that are now becoming critical such as analytics and unstructured data processing.
Flash, especially when used as a storage medium inside servers, presents far more possibilities to design efficient data centers than ever before. All-flash arrays (AFAs) are certainly useful, but in actual use they are mostly deployed for improving mode 1 data centers that already rely on shared storage via a SAN fabric. There is certainly nothing wrong with this, but it does not create a mode 2 approach.
The ‘serverfication’ of storage is upon us, combining flash with software-defined storage methods that effectively provide data management capabilities while leveraging the power of today’s server platforms. Especially when combined with PCI Express inside these servers, flash storage provides enormous ‘horsepower’ for previously I/O-bound workloads, and physically locates data close to the CPUs—very important for the reduction of latency and the subsequent reduction of runtime for these mode 2 workloads.
Finally, the use of flash provides outstanding total cost of ownership (TCO) for mode 2 data centers, effectively replacing many rotating hard disk drives (HDDs) for a single SSD in terms of the ability to service workloads. The TCO of using flash in mode 2 versus HDD in mode 1 is proven (see the Micron Cassandra and MongoDB case studies here).
There is an absolutely solid case for deploying flash in mode 2 data centers. I am blessed to be able to work with many customers who ‘get it’ already and are adopting server-based storage approaches using flash. For those who are not, the opportunity awaits. Start up your next project using a mode 2 approach—you won’t be sorry. In the long run, your data centers will be much more efficient with greatly reduced OPEX, and—most of all—your users will be delighted.
I’ll see you on the road somewhere! Until then, remember: Micron is flash, and flash is SOLID.
We Want to Hear From You!
Leave a comment below, or connect with us on Twitter @MicronStorage.