It has the geeky name of Non-Volatile Memory Express or NVMe, but the technology is changing how data centers operate, boosting performance and efficiency and transforming the capabilities of government agencies.
With near-constant change in data center technologies, it’s often difficult to separate incremental innovation from a once-in-a-decade sea change. The new NVMe (Non-Volatile Memory Express) protocol, its use in NVMe solid state drives (SSDs, sometimes referred to as “flash”) and its ability to be used over a network (NVMe over Fabrics) is the latter. This relatively new technology can help government IT deliver performance and efficiency improvements, as well as enhance storage utilization and ROI.
Knowing when NVMe SSDs are cost-justified compared to traditional SSDs in specific workflows and applications allows agencies to do more with the same budget and assure today’s data center infrastructure bridges smoothly to future computing needs.
NVMe is a standardized and simplified interface for SSDs that accelerates the transfer of data between computing resources and the drive. When directly connected to the computers’ central processing unit (CPU), with no intermediary “controller,” NVMe drives enable streamlined access to next-gen storage and are more suitable for massive performance and scalability. It enables IT teams to process more data faster, and the cost per unit of performance (as measured by cost per input/output operation per second, or cost per IOP) drops dramatically.
NVMe products began to appear on the market toward the end of 2014, and NVMe-leveraging storage platforms emerged in 2016, making it a fairly new, but stable, technology. Today, every major storage vendor has an NVMe offering.
The magnitude of NVMe’s impact on data center operations has been significant: massive IO parallelism, about six times the performance of comparable drives using the older Serial ATA (SATA) interface, up to 10 times lower latency, and better power efficiency. In fact, there is so much performance that with multiple NVMe drives, a single host struggles to utilize more than three to four drives. This led to the introduction of NVMe over Fabrics (NVMeoF), allowing for the NVMe protocol to go over a network, and letting multiple systems utilize remote NVMe drives.
What does this mean for government IT?
For big government agencies, NVMe’s ROI is highest when applications require both high throughput and low latency, as with analytics, decision support systems (DSSs) query, online transaction processing (OLTP) and graphics-intensive applications.
However, because NVMe media is more expensive, system architects often allocate it to specific applications or teams whose work demands it. While appearing cost-smart, organizations end up with “storage silos” with under-utilized NVMe Flash capacity.
By employing an NVMeoF, NVMe storage devices can communicate over a high-speed network enterprise-wide and thus can be virtualized and shared, so unused storage can be accessed for other workloads anywhere in the organization. Governmental IT departments can maximize how they use NVMe resources, support an agency’s demanding applications at scale and squeeze more ROI out of their expenditures.
Data warehouses are an example of where NVMeoF has a clear advantage over other approaches. Query speed determines what kind of response time end users experience, and as user counts grow, IT teams must be careful that storage resources can keep up the pace.
For example, in New York, the state’s central Medicaid Data Warehouse (MDW) enables over 3,000 users to simultaneously query Medicaid data to help improve care delivery for more than 6.6 million residents covered by Medicaid. With data from 10 billion Medicaid claims and encounters, MDW is massive and its growth, costs and response times were becoming unwieldy.
The New York State Department of Health modernized its warehouse with NVMe and has increased the percentage of ad hoc queries completed in under one minute to 98 percent, up from 50 percent. Because its NVMeoF implementation is software that runs on common, off-the-shelf servers, the state isn’t locked into purchasing only one type of hardware device from a specific vendor, helping future-proof its storage architecture.
NVMeoF also supports how an IT department can consolidate disparate storage devices it already has on site, generating significant savings depending on the circumstances. For example, a large health-care agency ran a health-care payment reimbursement analysis across two storage platforms — one of them with dozens of smaller databases for specific projects — because neither platform was capable of handling data volumes on its own. End-user performance suffered as it took too long to move the data from each resource to each individual database. Once moved, the agency’s queries took too long to run.
Moving to NVMeoF allowed the agency to operate two legacy storage resources with just one lower-cost system. The elimination of the legacy network between the two former storage resources increased end-user performance. The agency saved 50 percent on both capital costs of the storage hardware and operating expenses on software licenses and maintenance, and enabled the use of commodity hardware as newer, less costly storage media debuts.
The infrastructure at many agencies is a 40-year-old compendium of multiple hardware and software solutions. The path to modernization is complicated. Present performance can be inconsistent, and the need for technical staff to maintain a diverse set of resources can involve ongoing consulting needs and high personnel costs. Technological advances in NVMe storage and NVMe over Fabrics help address common pain points for government data centers, with often compelling economics and price/performance.
Brian Dougherty is chief technology officer at CMA, a major government system integrator based in Latham, N.Y., which provides the Mosaic SSD NVMe database platform.
Josh Goldenhar is vice president of products at Excelero, which provides the NVMesh® software-defined block storage for shared NVMe at local performance.