Issue link: http://cp.revolio.com/i/833286
8 | THE DATA CENTER JOURNAL www.datacenterjournal.com IT fortress oen provides a false sense of security. An obvious solution to moving beyond the IT fortress is the cloud, but pure virtualized cash-and-carry style operations have limits as well as benefits. Firing up a single virtual machine requires a credit card, a cloud provider and five minutes or less of setup. It's easy to do, but how well such public "drive-in" services can scale to meet the needs of machine-learning projects is uncertain—especially those projects requiring fast silicon. Even if they can meet those requirements from a pure specification standpoint and provide service-level agreements (SLAs) that ensure the success of a compute-intensive project, they can- not do so cost-effectively when scale is required. As companies rely more on machine learning as an integral part of their compute requirements, forward- thinking CIOs must think not only of today's need but also of the power, space and cost requirements over the next decade. Organizations that are expanding their machine-learning capabilities should be wary of lock-in and keep their options open. Colocation gives companies the ability to scale beyond the cloud while maintaining control of their costs without the overhead of staging and hosting them in a traditional IT fortress. Depending on the required services and where the facility is located, colocation can provide up to 80 percent savings over a traditional in-house IT fortress. Today, you can lease turnkey physically secure data center space, fully equipped with power and cool- ing, and have the ability to start at a single rack of equipment and scale to multi-megawatt solutions. In this space, you have control of how you want to architect computing re- sources, and you have the confidence of knowing where your data resides— unlike a cloud "black box" virtual solution, which is unlikely to be pow- erful enough for a dedicated machine- learning project. And by choosing the infrastructure carefully, IT leaders can retain the ability to burst into public- cloud environments, preserving the ability to expand the business regard- less of the changes ahead. Selecting the right colocation service provider is important. Coloca- tion physical security should be equal to or better than that of any corporate data center. Standard features should include a separate security lobby, camera surveillance with 24x7x365 monitoring and card-key access to secured areas, along with standard fire protection and electrical-power-sys- tem monitoring. Depending on geo- graphical location, operational costs can be lower thanks to the availability of free cooling rather than a need for more-elaborate HVAC systems and the power they require. Power may be the most critical factor in selecting the appropriate co- location, with reliability and cost both playing important roles. A well-engi- neered colocation facility should have engine-generator backup, UPS power backup and redundancy for electrical and mechanical cooling systems. But colocation also provides the unique opportunity to shop for operational savings on the basis of the electricity price—something you just can't do with the traditional IT fortress. In the larger world of corporate IT, power is a major part of monthly operating costs for any compute- intensive project. Perhaps the greatest benefit of distributing projects beyond the corporate IT fortress is the ability to shop for the lowest cost per kilowatt and to lock in low-cost electricity. For instance, colocation providers in Ice- land can offer power costs that are the lowest in Europe, with pricing protec- tion to the year 2030 and beyond. is feat is possible because the Iceland power grid has completely eliminated its dependency on fossil fuels. e cost of electricity need not fluctuate, as the source is hydroelectric, geother- mal and wind energy. e benefits of long-term, low- cost electricity pricing will be impor- tant to companies that are adopting machine learning and other resource- intensive applications. Even if a traditional IT fortress has the physical infrastructure to support a rapidly expanding project, it oen consumes a compute cluster of more than 10 racks of servers and GPU accelera- tors for days, if not weeks, requiring more than 300kW of power. e CFO may be uncomfortable with watch- ing the facilities budget go through hockey-stick growth as servers are turned up. It's a "gotcha" that need not happen when you can shop for power by choosing among colocation data centers located in regions with access to low-cost power generation. Fear drove companies to build out their IT fortresses. Now, competi- tive fear is driving organizations to think beyond the fortress. Forward- thinking executives are pushing the boundaries of the fortress to include access to cloud and colocation solu- tions that sit beyond the traditional walls of the IT fortress. Machine learning and other high-compute/ high-resource activities will drive this adoption. To stay ahead, the leaders will use the latest powerful and cost- effective data center facilities to ensure that they have access to the latest in technology, all while driving growth at the bottom line of the balance sheet. n about the author: Tate Cantrell is Chief Technology Officer at Verne Global.