Never an Island: The Importance of High Bandwidth in a Data Center
The rapid growth of cloud computing and the ramping-up for 5G services is driving massive demand for high-capacity network and data center infrastructure. Both require the ability to deliver huge amounts of data reliably at extremely high-speeds to ensure uninterrupted transfer between the data centers and the end user. With the demand for data and the volume of data continuing to grow exponentially, compute and storage of the data center will rely on the networking infrastructure that meets latency, jitter and performance requirements now, and in the future.
Data center and transport network services are both links in the IT chain, no data center can function as an island and without compute and storage, networks have limited usefulness Data and the flow of information need to be both secure, accessible and able to easily and reliably move large quantities of information at speeds greater than human patience. The synergies between a network provider and the data center can greatly improve the level of responsiveness and satisfaction a business looks for when choosing a data center.
Size and Scale
The sheer scale of today’s compute and storage demands a resilient, high-capacity transport network that can meet business needs. A 2018 report from UntiTrends shows that a full 27% of respondents reported having over 100TB of data on back up, more than double the percentage from 2016. Correspondingly, there is a significant decline in the percentage of respondents that protect volumes under 25TB. Businesses continually evaluate new storage technologies and methods to ensure the growing amounts of business data is protected from any failure. The entire data storage, protection and business continuity process will be more involved, more complex, and costly if the chosen data center cannot provide adequate redundancies and sufficient transport bandwidth to support these operations.
Industries such as financial services, media, e-health, and retail, enhanced by 5G mobility are not only transferring and storing more information, but are also coming to expect diminishing application latencies, i.e., faster networks. According to Light into Money: The Future of Fibre Optics in the Data Centre Networks by Dr. Rick Pimpinella, 5G networks upload speeds will be able to match the download speeds of today’s 4G LTE networks. 5G download speeds will be in the 1-10Gbps range, with uploads as fast as 1Gbps. Today’s data center networks must be prepared to transport the volumes of data that will be generated and provide the 1 millisecond latency throughput required.
Secure and Accessible
According to a 2018 report from UntiTrends, over 30% of companies reported data loss three years in a row, from 2016 to 2018. More companies are beginning to realize the importance of off-site storage and are deploying secondary sites for disaster recovery even if the application lives in the cloud. Over 40% of companies in 2018 reported use of a secondary disaster recovery site and over 75% of companies perform disaster recovery checks on at least a yearly basis. While this increased reliance on off-site backups is encouraging for industry and reassuring to their customers, and effective disaster recovery plan places even more emphasis on dependence of high speed network access. Theoretically, a data center operating at only 10 Gbps can transfer 2 GB of data in about 1.6 seconds. While this may seem impressive at first glance to an outside observer, the digital world operates not in seconds, but milliseconds. Increasing the speed ten-fold to 100 Gbps and a data transfer can still take nearly 160 milliseconds. Only at speeds of 400 Gbps, will a transaction start to resemble ‘real time’, at a mere 40 milliseconds. The higher speed transfer enables an organization to mirror its transactions and be confident of almost full recovery in the event of disaster.
Off-site backups are always going to take longer to download and implement because their bandwidth speeds will never be able to compare to a local backup. This makes a high speed network access even more essential in the event of data corruption, image corruption, malware attacks, network or infrastructure failure or natural disaster for on-site storage. Companies must know that their data is quickly and easily accessible and secure, both within the data center and across the network.
What to Look for
There are several ways to address the need for higher bandwidth in data centers. Dr. Pimpinella places responsibility solely on network architecture technology: “At the transport layer, there is unlikely to be a single answer to networking when it comes to future proofing an investment. But the velocity of growth for speed and capacity is only going one way and the networking technology which companies trust to scale linearly is Ethernet on single-mode and multimode fiber optic cable”.
Pimpinella also recommends a change in network topologies, altering the physical layout of the fiber itself. He encourages the move from hierarchical three tier architectures to a single non-blocking switch or leaf spine architecture: “Hierarchical three tier architectures will have far too much latency to accommodate AI and ML… In leaf spine architecture, every leaf is connected to every spine but the spines are not connected. This alleviates oversubscription and latency problems”.
Marvel Technology Group focuses on the interplay between a data center and its network as important factors for keeping up with high data bandwidth demand, citing the ability to bundle and unbundle channels and growing demand for higher-density switches, and also the increased importance of power efficiency to keep costs down. Data center design has always been about efficiency and this is doubly important when managing the interplay between the physical environment of the data center space and the multiple secure, high speed network paths. If network and data centers are managed under the same company, any improvements that can be made on the data center side will have even greater impact on network performance.