As Storage Switzerland previously blogged, non-volatile memory express (NVMe) access protocols stand to add value to a host of applications, ranging from the performance-intensive newer workload set (e.g. artificial intelligence) to more traditional Tier 1 applications (e.g. Microsoft SQL Server).
For many storage managers, the question becomes how to introduce NVMe to the data center in a way that maximizes the return on this premium-priced investment. In addition to maximizing utilization of available throughput and input/output operations per second (IOPS), maximizing ROI largely comes down to avoiding introducing a new silo or “island” into what are already fragmented and difficult to manage storage infrastructures.
To deliver extreme levels of performance, many storage vendors invest in custom componentry such as field programmable gate arrays (FPGAs) and application-specific integrated circuits (ASICs). This approach stands to drive up costs and complexities, and to limit the ability to bring a broader ecosystem of workloads to NVMe. Additionally, many early-to-market NVMe solutions have come to market with a limited, immature software stack – thus pushing the bulk of data management back on the application, and consequently limiting the realization of available performance acceleration.
The majority of storage managers should look for an NVMe architecture that uses standardized components for interoperability with existing data center infrastructure, and that supports multiple access protocols, including serial-attached SCSI (SAS). Providing pools of storage over a shared network can substantially drive down network latency, provided that workloads are tiered intelligently according to their performance and capacity requirements. A shared architecture can also increase visibility into storage resources – creating the opportunity for increased system utilization, as well as for more informed capacity planning.
Data management software is important to optimally balancing performance, cost and data availability in a shared NVMe architecture. Capabilities such as automated data placement, inline deduplication and compression can help to enable flash storage resources to be utilized more cost efficiently, while multi-parity RAID, asynchronous replication and encryption can help to make sure that the data feeding mission-critical applications is secure and available when it is needed. Data availability becomes a particular concern when undergoing any storage consolidation initiative, as the burden of serving a larger pool of mission-critical data falls to a smaller number of systems (or to a singular system).