By Sanjay Motwani, Regional Director, Raritan Asia Pacific, Raritan APAC
The data center, as we know it, has changed. There is a sea change in how organisations are using and valuing technologies such as wearable technology, Big Data, Internet of Things (IoT) etc. and this is prompting them to shift their thinking in terms of how they plan and build their data centers. Earlier, data centers focused on storage of information and disaster recovery. Now, it has shifted to analyzing and processing data for on-demand access. New technologies like mobility and wearables are demanding increasing latency.
Approximately 80% of all datacenter space (by square footage) globally was owned by enterprises in 2015. However, that percentage is expected to drop below 75% by 2020. The main reason for this shift is the migration of certain workloads from enterprise-owned sites into typically more cost-efficient cloud and co-location facilities. However, there will still be a requirement for dedicated, premium enterprise sites among some organizations. New edge micro-datacenters will also to some extent replace ‘legacy edge’ enterprise server closets and rooms, as well as supporting new use cases. Edge computing can be described as the distribution of compute and storage capabilities to the very edge of the network near the point of data generation and data use. This could be an enterprise factory floor or a carrier point of presence, a cell tower or a smart building.
Factors driving change in data center design, operation
Changes in datacenter design and operation are to some extent shaped by a number of underlying forces.
The forces of change are:
Demand: There are significant concerns that datacenter space, power and bandwidth may struggle to keep up with the explosive global demand for IT services expected over the next two decades.
Convergence: There are indications that the typically separate IT and facilities (equipment and staff) organizations are becoming more integrated.
Cost Transparency: Colocation and service providers are leading the way in terms of investing in technology to provide more transparency (e.g., power usage and IT capacity), showback and real-time costing.
Industrialization: It is likely that the massive scale of the global datacenter market will enable suppliers to produce or pre-configure an extensive catalog of equipment and designs, each optimized for customer-specific requirements and applications.
R&D: Research with a scientific imperative, but sometimes without a clear business case, will also shape future datacenter design and operation.
Accurately predicting the evolution of physical datacenter design and operation over the long term is obviously challenging. It is possible to identify a number of future datacenter types (some of which exist today) that are likely to dominate in the next decade and beyond and they include Hyperscale (cloud operators but also some service providers); Cloud (non-hyperscale) and service provider; Colocation (MTDC) and service provider; Enterprise (dedicated, premiumsites and fewer closets/rooms); Edge (micro-datacenters as well as core sites); and HPC and specialized.
Specific attributes & criteria defining the data centers
The data center types are defined by some of the following criteria and attributes:
Business Models: Business models will vary between datacenter types based on ownership. Some enterprises will have specific workloads, data requirements or governance issues that dictate that they must continue to design and operate their own premium datacenters. Given this, these organizations then have the opportunity to innovate and customize in ways that may not be open to commercial (colo, hosting) operators.
Scale: Despite advances in compute capacity and more workloads moving to the edge, there will be a continued and sustained need for hyperscales in the future. Small-scale micro-sites are expected to become more widespread to support IoT and other applications.
Resiliency: Resiliency requirement will be even more closely related to business case and function. There will also be greater adoption of software-based, distributed resiliency with less reliance on physical infrastructure (generators, UPS).
Efficiency: Some facilities, such as hyperscale sites, will heavily prioritize efficiencyabove most other criteria. A percentage of hyperscale sites will also continue to focus on sustainability and carbon reduction by utilizing more renewable energy.
IT Density: Average rack power density, currently less than 5kW, is likely to continue increasing over time, driven by applications such as AI/machine learning, high-performance computing and big data. However, some wild-card technologies (e.g., quantum computing) have the potential to increase compute capacity while significantly reducing power requirements. Density will increasingly be tied to business function and workload.
Geography and Distribution: A number of large cloud operators have built out in specific locations (e.g., Europe) or have leased from MTDC providers, partly in order to comply with data regulations. This trend is likely to continue for future datacenter types. Hyperscale sites will alsocontinue to be built in areas with low energy costs, tax incentives and climates that allow for free-air cooling. Edge capacity will be added in centralized datacenters as well as in metro sites outside of core datacenter hubs.
Why move to the Edge
Of all the trends shaping future datacenter development, the demand for edge computing is expected to be one of the more significant and, as such, deserves specific attention.
While use cases such as content distribution networks (CDNs) and local processing and storage are expected to drive edge compute demand in the short term, IoT is expected to be one of the long-term drivers for new edge capacity. IoT’s uses cases are vast but even within similar use cases, data paths and datacenter types will vary.
It does seem likely that a number of IoT deployments will end up having data residing in a combination of public cloud and non-public cloud facilities, with the need for both distributed micro data centers and very large centralized sites.
Key data, including data needed by other applications and people, will in some cases be made available at the ‘near edge’ in large data centers where colocation and other metro data centers are sited close to where the data is generated. Cloud heavyweights are rapidly building hyperscale data centers with direct fiber links to leased colocation sites. This brings hyperscale cloud capacity closer to the edge – effectively functioning as ‘near edge’ datacenter capacity. Cloud providers will also utilize ‘cloudlets’ – distributed edge capacity for data caching or low-latency compute.
Once consumed or integrated, data will then typically be moved or streamed into large or hyperscale remote datacenters to be aggregated, analyzed (including through integration with other data and applications) and archived. These large facilities represent the ‘core layer.’
As the edge becomes more intelligent, legacy physical hardware will be more expensive to maintain.What’s more, legacy gear won’t be able to offer the data and traffic flow capabilities that are required of intelligent infrastructure.Network abstraction gives IT managers more centralized control rather than having to maintain an operating system on each and every device, while also performing faster rollouts and upgrades. There are several tools and softwares such as Power IQ, iX7™ controller-based intelligent power distribution units (iPDUs), Intelligent Cabinet prototype and Dominion® SX II that can help IT managers in better provisioning leading to higher resiliency, stability and reliability in the emerging hybrid architectures with a mix of core and edge DCs. Trends like IoT will lead to the generation of massive volumes of data, while companies continually seek declining levels of latency. By moving analytics to the edge, organisations will get more immediate insights and results from IoT devices and sensors. This will only encourage the need for edge data centres to exist alongside their larger, centralised counterparts.