Meeting Hyperscale Compute Needs with Non-Volatile Memory

Bookmark and Share

We live in a real-time world where no one likes to wait. Incessant demand for faster end user experiences and services, on every mobile device creates unrelenting pressure on the server side of the internet, where data centers power our daily lives. As digital content reigns supreme for these global internet platform providers, companies must serve customers quickly and economically.

“Hyperscale” is a computing term for software-centric architectures that scale appropriately upon increased system demand. These data-driven environments must scale predictably and efficiently through seamless integration compute, memory, networking, and storage resources. Hyperscale computing is how large, distributed sites such as Facebook, Google, or Amazon build reliable and scalable systems infrastructure to serve billions of users. (1)

Companies deploying hyperscale architectures serve the widest customer segments, build the largest data centers, and break new ground with real-time, data-centric services. To do so, they require hundreds of petabytes of server-side non-volatile memory to effectively provide the services that better the lives of their global customers.

Showcasing the breadth of hyperscale’s impact on our daily lives, a broad group of hyperscale providers was listed in a recent presentation on The Personal Revolution and The Data Factory by Michael Moritz of Sequoia Capital. Moritz’s talk included examples of marquee consumer technology brands, including Apple, Facebook, Amazon, Google,, LinkedIn, eBay, Netflix, Twitter, Spotify, Airbnb, and Unity. The presentation also listed platform providers in emerging markets, including Tencent, Alibaba, Baidu,, Ctrip,, Yandex, Naver, Y! Japan, Rakuten, and Asos. (2)

The Hyperscale Market is Beginning to Drive Transformations

The hyperscale market is just beginning to drive major transformations in compute infrastructure. Leading platform providers like those listed by Moritz have identified unexploited computing avenues to deliver instant, rich, immersive digital experiences to a global audience. Their flexible, software defined infrastructure relies on the benefits of flash memory, which is quickly transforming data center infrastructure in both hyperscale and enterprise environments.

Specific Benefits of Hyperscale Flash Memory Deployments

Flash, implemented inside the server, supports exponentially more transactions per gigabyte (GB), at lower latency than disk drives, which rebalances CPU and storage performance for the first time in many years. This equates to maximum infrastructure efficiency, which enables hyperscale companies to focus on software to scale and tune the datacenter.

Flash Provides More Transactions per Kilowatt

For hyperscale companies, power consumption is one of the most crucial elements driving data center costs. More transactions and workloads performed per kilowatt (kW) means lower operating expense per square foot of datacenter.

By optimizing transactions-per-kW, hyperscale leaders deliver more for less. This aggressive change to the memory and storage equation with non-volatile memory, rather than a traditional dollars-per-gigabyte assessment, drives immediate innovation and economic return.

Flash Enables More CPU Workload per Cubic Centimeter

Maximizing workload per cubic centimeter of precious datacenter space allows companies to better build green data centers while still serving billions of customers faster than ever before.

Similar to reducing power consumption, given the radically lower profile that flash memory provides, non-volatile memories also reduce physical space requirements. Similar to smaller, thinner flash-powered computers, tablets and mobile devices, shrinking the space profile for hyperscale server platforms means smaller, more efficient data centers. In tech, smaller is almost always better, both for cost reasons and user experience.

Server Side Flash Storage Equates to Vast Efficiency Gains

The shift from mechanical disk drives to intelligent, persistent silicon media equates to vast efficiency improvements. Today, hyperscale companies leverage flash storage inside the server to balance CPU and storage performance while maximizing the utilization of costly datacenter resources. Reliability and performance features are integrated into the server layer with intelligent algorithms, eliminating vast storage infrastructure that would have been unmanageable, costly and power starved at hyperscale speeds.

Hyperscale Leaders See Savings from Infrastructure Innovations

Hyperscale leaders are seeing  savings from their infrastructure innovations. In a recent white paper, titled “A Focus on Efficiency,” Facebook detailed how it has achieved power usage effectiveness ratios of as low as 1.04 in an industry accustomed to ranges more like 1.5 to 1.9. (3)

Spotify, in architecting its Cassandra cluster, found that non-volatile flash memory “gives us the speed and scalability we need to grow our footprint worldwide with new services and scale our user base by the millions.” (4)

Like Facebook and Spotify, there are many areas where hyperscale companies can benefit from new architectural models based on non-volatile memory. We have moved far beyond “if” to “when and where” these new media types will be deployed. Fortunately, there are plenty of options to choose from as we race toward the all-flash data center of the future.


(2) The Personal Revolution and The Data Factory by Michael Moritz, Sequoia Capital,



Image of network server room with racks courtesy of Shutterstock.