The IT industry is characterized by loops of iterative innovation – hence such adages as "back to the mainframe again". However, recent reports regarding the potential future design of datacenters from Emerson Network Power – Data Center 2025: Exploring the Possibilities – really outline the state of the industry in relation to drivers from cloud technologies on the inside of the business as well as externally as public cloud providers.
The report highlights differing opinions naturally, however, there are some rather radical thought paradigms in place that may well entail a whole new way of internal IT thinking on behalf of the CIO and infrastructure management:
43% likelihood of Private Power Generation in Hyperscale Data Centers in 2025
Average power density of data center expected to reach 52kW/rack
67% of survey participants believe at least 60% of computing will be cloud-based by 2025
Growth of cloud will ultimately result in the decline of the 1-2MW enterprise data center
58% believe that data centers will be 50% smaller than current facilities
41% see a combination of air and liquid being used for thermal management
Data centers will be highly automated – all devices auto-identify and operated without/minimum human resources with utilization rates expected of 60-80%
- 50% expect server lifecycles to stay in 3-6year range (as today) with Asia Pacific projecting 7+ years
Figure 1: Overall Response from Participants
In short, some very serious implications for anyone that is running a private datacenter today. The relatively conservative pace of IT innovation within the datacenter is set to change based on new computational and storage needs with their implicit effect on networking and physical infrastructure.
In the same report, Andy Lawrence, vice president at 451 Research summarizes nicely by saying:
"The data center of 2025 certainly won't be one data center. The analogy I like to use is to transport. On the road, we see sports cars and family cars; we see buses and we see trucks. They have different kinds of engines, different types of seating and different characteristics in terms of energy consumptions and reliability. We are going to see something similar to that in the data center world. In fact that is already happening, and I expect it to continue."
Datacenter management needs to consider how such a metaphor can be instantiated within their own "bricks and mortar stores" to be enable a bridge to the future right-sizing of datacenters taking place.
Opportunities for the CIO
Even within the confines of this report and it predictions, there are areas of innovation to still explore. The need to have smaller more ultra-dense datacenters provides the following advantages for enterprises to utilize fast nimble techniques that are simply not economical for a large public cloud provider.
Service quality excellence
In a world of "public" public cloud downtimes announced within a fraction of a second by those relying on these constructs through social networks, service quality excellence and superior uptime remain the hall mark by which business will measure the value of internal IT.
Vertical capability integration
With much of the heavy network and storage traffic being East-West in a classic datacenter, vertical capability-focused integration (as opposed to simple infrastructure integration) will provide ultra-dense configurations handling superior numbers of concurrent workloads whilst maintaining service quality excellence. Typical workloads such as database would be able to take advantage of much faster networking (typically Infiniband 40-56Gbps vs 10GbE in North-South directions) with local modular storage and usually results in massive reductions of footprint, licensing and facility needs. Interlinkage with other capabilities (such as middleware or analytics) can be facilitated over the same high-speed fabric rather than inefficiently "hop-scotching" over the entire classic datacenter.
"Disposable" Private Cloud Datacenter Containers
Ultra-dense capability computing will again be a workable idea using the state of the art in air/liquid cooling. These can basically strip out the entire traditional costs of building out facilities and are simply more flexible for location. Further, they can utilize low factor costs in geographies where a traditional datacenter can't be built without extreme cost being incurred. Full automation of the container allows a notion of "decentralized centralized IT" to still be put in place. The idea of a "disposable" container – literally swapping in/out as newer technology merits has also been bandied about as a way of extracting micro-location financial yields through asset sweating.
Urban Datacenter Grids & Locality
Datacenter containers allow enterprises to physically locate "their" private clouds close to urban business concentrations allowing superior service quality, low latency and ability to differentiate service offerings at software and hardware level. Trends in creating datacenter container parks may well arise that allow a further sharing of common estate costs.
- Intra-Inter Rack Form Factor Design
Ultra-density imposes challenges on form factors of traditional rack-and-stack configurations. Many initiatives are underway to further optimize within the rack itself using rack-level disaggregation combined with a rack-level component fabric. Intel's Rack Scale Architecture (RSA) is currently gaining most media attention, but is by no means the first. Indeed Oracle Engineered Systems are architected in a similar manner as are appliances from Teradata and others.
The CIO of today is constantly bombarded with messages from the business and the press about moving all to the "Cloud". However, the potential for enterprise IT innovation remains essentially quite strong.
Complete and realistic business case calculation of public cloud providers, including opex costs for storage, compute and importantly network can be used to justify and finance an innovative cost-effective/cost-neutral enterprise datacenter strategy with much higher standards of security and service quality. The public cloud is providing a generational jump from IaaS in the forms of PaaS and functional SaaS capabilities. However, they do not cater for all the security and uptime quality needs of all enterprises.
To differentiate and capitalize on existing talent and resources, the CIO may need to buck the trend in keeping a fleet of smaller enterprise datacenter containers that are used to extend the "commodity" functionality in the Public Cloud PaaS and SaaS to in/on premise IT as a means of innovating beyond the cycle of the Public Cloud industry.
After all – if everyone does the same thing – then we should expect a similar level of result albeit more efficient than today!