Many new data centers being built today are being located at greater distances from typical enterprise users than in the past. Content data centers initially were built in and around metro core areas to provide the most effortless access to content by corporate users. Given the rising cost of real estate, power and cooling, enterprises are situating new data center developments farther away from metro core areas to reduce their operational costs and enhance geographic resiliency. The fundamental need for users to access content has not changed, but the distance to access that content has grown from metro distances to regional distances.
For example, companies such as Amazon, Dell, Google, Microsoft and Yahoo have located their large-content data centers in Washington State. Many of these facilities are located in the rural town of Quincy, which has access to several hydroelectric dams delivering low-cost green power. Similarly, Amazon built a data center in Umtilla, Washington, to access low-cost hydroelectric power. These rural locations are up to 300km away from the closest regional points of presence in Seattle and Portland, and like other remote data centers, their location creates a number of new challenges, including the following.
1. Increased inter-metro bandwidth requirements
The rapid growth in content is driving many large enterprise data centers to be interconnected across metro-regional boundaries at very high capacity, with the bandwidth requirements consistently in multiple 10G, 40G or even 100G Ethernet services.Using a traditional metro deployment model, providing this high capacity across metro-regional boundaries is very expensive and slow to implement. With the move to more-remote data centers, this traffic is increasingly traveling outside the traditional metro network domain and into regional and long-haul networks. In addition, enterprises need on-demand connectivity to address the on-demand cloud-comsumption needs of customers.
2. Challenges around crossing network boundaries
When enterprise data centers were built exclusively in metro areas, connectivity between the data centers would span perhaps 50km and stay within traditional metro network boundaries. The remote data center trend shows a number of regional networks connected via a backbone network, where connectivity now must span three different networks (the originating metro, the regional network and the terminating metro network). This is not a minor detail, and the result is network inefficiencies as a result of having multiple network handoffs, increased provisioning and engineering effort for new services, and a complicated process to manage end-to-end service performance for the enterprise customer.
3. Outdated and inefficient data center connect models
Many service providers have realized that they need to transform the way they build networks to connect end users to data center content, evolving toward a ”user-to-content” model. By blurring the boundaries of metro-regional networks using this user-to-content approach, service providers can cost-effectively support the expected traffic growth of multiple 10G and 100G services and beyondwhile building a foundation for new, variable-bandwidth services.
Evolving to the User-to-Content Model
Although there are common challenges in the trend of moving data centers to more-remote locations, there isn’t a one-size-fits-all solution for connecting them. Here are several distinct scenarios for enterprise traffic demand to the data center, along with a suggested approach for handling each situation with maximum efficiency and return on investment.
1. High-static demand where large enterprise data centers have the need to be interconnected across metro-regional boundaries at very high capacity, and traffic represents greater than 50 percent of a 40G or 100G wavelength.
Many large enterprise data centers need to be interconnected across metro-regional boundaries at very high capacity, with the bandwidth requirements consistently in multiple 10G, 40G or even 100G services. Using a traditional metro deployment model, providing this high capacity across metro-regional boundaries is very expensive and slow to implement. The alternative is to connect directly the two data centers across metro-regional boundaries, creating an “express” wavelength, which in fact blurs metro-regional boundaries at the optical layer.
2. Low-static demand where it isn’t cost-effective to design express wavelengths for each individual customer demand, and traffic is less than 50 percent of a 40G or 100G wavelength.
Not all enterprise data center sites require the high-static demand of multiple 10G described in the previous example. For lower-demand sites, it is not cost-effective to design express wavelengths for each individual customer demand. A more cost-effective approach is for traffic from multiple low-demand static sites to be aggregated using a hub and spoke model into a single high-capacity 40/100G link into the regional backbone.
3. Variable demand where traffic is required on scheduled, in short-term intervals.
An example of this type of demand is the regular data backup that IT organizations perform over the weekend. To maximize application performance, enterprises would need to procure a 10G service for three hours every Saturday night. With static service offerings, enterprises would usually buy service capacity that is typically required for day-to-day operations, such as a 1G service. As a result, a backup operation using static services would require a much longer duration. Other applications might include spikes in demand for distributing media content; for example, they would require 10G services to distribute high-definition preproduction, uncompressed video streams from event venues to production studios. On-demand services are delivered on OTN/packet switching network architectures by adding software-defined-networking-type (SDN-type) capabilities.
Although traffic and technology are evolving in the metro network, the industry’s handling of bandwidth growth has not. Network operators still commonly employ the traditional method of deploying more and more routers throughout the metro network. Because Ethernet is fast becoming the dominant technology both inside data centers and in the metro networks that interconnect them, it makes sense to deploy networks that are raw and powerful, are able to switch and aggregate Ethernet, and reduce the cost and complexity of deploying high-touch functions that are unneeded in many parts of the metro network.
About the Author
Mitch Simcoe is a director of industry marketing at Ciena. An industry veteran with more than 25 years of telecommunications industry experience, Mitch is responsible for promoting Ciena’s packet optical solutions for metro-regional networks. Before joining Ciena, Mitch was director of application marketing at Genband, where he built an application showcase for communication-enabled IT applications. While at Nortel, Mitch held numerous roles in sales, marketing and product management, with his expertise spanning voice switching, broadband access and optical networking. A graduate of McGill University in Montreal, Canada, Mitch holds a Bachelor of Engineering (electrical) and a MBA (marketing).