In today's business climate, the need for new IT infrastructure is relentless, as is the pressure on enterprises to reduce TCO and improve efficiency. Even increased datacenter spending will not deliver the capacity you need with the current ratio between capacity and cost.

At the same time, we have to embrace the vast opportunity of a hyperscale future, one that goes beyond the limitations of today's architecture—making it smarter and more modular and flexible. This calls for nothing less than a fundamental rethinking of the very nature of the datacenter.

The limitations of today's hardware-defined infrastructure

The low level of resource utilization in datacenters today is partly caused by resource stranding: a fragmentation problem created by varying application profiles (such as compute-intensive, memory-intensive, and networking-intensive applications). To address this challenge, datacenter operators employ virtualization technologies such as hypervisors and containers but, even so, they cannot overcome the hardware-defined infrastructure boundaries.

This hardware setup is a result of the monolithic nature of component integration, leading to a one-to-one mapping between a physical server chassis and a system executing on that chassis (which is referred to here as the host; see the figure below). Hosts, therefore, have fixed configuration properties, and this makes it extremely difficult for them to adapt to varying applications because, in most cases, the entire physical server chassis needs to be planned for each application profile.


Current hardware component integration also tightly binds datacenter lifecycle management to the lifecycle of an entire server. This causes problems for providers who wish to upgrade only part of their infrastructure, because different resources within a server might have different lifecycles.

Software-defined infrastructure – a first step towards hyperscale

Researchers and information and communications technology (ICT) companies have been working toward realizing hyperscale by focusing on the principles of hardware disaggregation and programmable infrastructure. With this approach, physical server limitations are removed, and resources (for example, compute, memory, and storage resources) are considered as individual, modular components.

This software-defined infrastructure approach breaks the existing one-to-one mapping between a physical server chassis and a host executing on that chassis, as depicted in the figure below. In this context, host composition is done independently from hardware-specific dependencies, allowing datacenter operators to employ resources in more efficient ways (in terms of increased utilization and lower power consumption, for example).


Hyperscale datacenter-driven use cases

In a flexible and modular datacenter environment, numerous use cases can arise, including individual component replacement, flexible host setup, and dynamically optimized datacenters.

If you want to explore the key considerations and questions for datacenter operators at every layer—from hardware to applications—please read our new white paper: Hyperscale cloud—reimagining data centers from hardware to applications.

Download the white paper

Cloud Infrastructure

Jesper Tunér

Jesper Tunér joined Ericsson in 2006, after working at Accenture as a management consultant and, prior to that, at a small software company as a programmer. He has held several international positions in marketing, sales and finance throughout Europe, Asia, Africa and the Americas. In 2013 he was appointed Head of the Cloud Marketing Program at Ericsson. Since the start of 2016, he has worked with Cloud Strategy & Portfolio Management for the Product Area Cloud Systems in the Cloud & IP Business Unit. Jesper has a BSc in Computer Science and a MSc in Business Administration from Lund University, Sweden, followed by post graduate courses at IMD Business School and Kellogg School of Management. He is passionate about the digital industrialization opportunities that come with cloud technology.

Jesper Tunér