As emerging technologies stretch data center bandwidth needs to the limit and beyond, enterprises are racing to deploy 100G and 400G technologies to keep pace with growing network user demands.
Yet as managers strive to maintain essential network services, it's become all too easy—especially for small and medium-sized organizations—to get caught up in the hype and waste money on bandwidth they really don't need. "The biggest mistake managers make when estimating bandwidth requirements is not aligning with the company’s strategic and long-term business objectives, which can leave them undersized," warned Trent Blakely, senior product manager, global managed network services, at IT protection and recovery services provider Sungard AS. "Having a good understanding of long-term business objectives, such as expanded growth in new market sectors or geographies, upcoming mergers and acquisitions, or cloud initiatives, will help managers properly plan and forecast bandwidth requirements," he noted.
Metrics reveal needs
A core requirement for accurately estimating bandwidth needs is establishing a system for monitoring and tracking network metrics. "It's from this baseline data that all of the opportunities and benefits can be realized," said Chris Haun, lead network architect at ServerCentral Turing Group, a managed IT infrastructure service provider. "Don’t overthink this; just get the baseline data and then start building from there."
Managers who fail to monitor network metrics risk losing both money and network reliability. "We often see people not running even basic bandwidth monitoring, which should never happen," Haun observed. He added that there are free and reliable open-source tools, such as MRTG and Cacti, that are easy to set up and provide valuable insights into bandwidth utilization. "We typically recommend moving up to the next size circuit once you’re hitting 50% of your capacity on a regular basis," he advised.
Regardless of a network's size and configuration, bandwidth requirements are largely driven by applications, the number of internal/external systems connected to the network, and how many end users are accessing applications, Blakely explained. Bandwidth needs can be further broken down by the internal requirements needed to support LANs and SANs, and by the external demands imposed by WANs and the Internet.
When estimating LAN bandwidth, it's necessary to decipher existing application workflows and system interdependencies in order to recognize the throughput requirements necessary for different types of traffic. "Understanding traffic flow through other LAN devices, such as firewalls, load balancers, IPSs, and WAFs, is a key consideration to make sure each component supports overall throughput requirements," Blakley said. For WAN connectivity, bandwidth is primarily driven by the total number of remote sites and end-users, as well as application requirements. When estimating WAN bandwidth needs, managers should consider several key factors, including voice and video demands, local or centralized Internet access, and MPLS or SD-WAN service requirements.
Chris Reidy, vice president of IT operations for car shopping service CarGurus, believes that the best way to get a firm grasp on how much bandwidth a data center needs is to monitor and record traffic growth and then use a calculated projected growth figure to plan for future bandwidth needs. It's also important to plan for enough bandwidth to support occasional traffic spikes. "At CarGurus we're able to be prepared for 5x traffic spikes to our site," he said.
Haun recommended creating alerts that will automatically generate a warning whenever a committed data rate is exceeded for more than 15 minutes. "This increment will help you see whether it's a one-off spike or something more significant," he explained.
Perhaps the biggest mistake managers make when estimating bandwidth requirements is failing to prepare for traffic spikes and not allocating enough bandwidth for the redundant links that become service lifesavers when the primary link fails. "The time spent and relatively small expense of having surplus bandwidth far outweigh the ramifications of significant slowdowns or, in the worst case, the site not responding," Reidy noted.
There are really no shortcuts to getting data center bandwidth right, observed Josh Williams, vice president of solutions engineering for INAP, a data center, and cloud service provider. "Most resource-strapped businesses, unfortunately, do not have the time, people, or tools to make 100 percent accurate assessments, especially in the case of new applications coming online," he said. "If working with an existing environment that has already gone through a 'guess and see' approach, managers at the very least have an historical baseline to measure against."
With bandwidth costs rapidly plunging, it really makes little sense to be too frugal. "In general, IT managers should always remember that bandwidth is a small expense to pay compared to a bad user experience," Reidy warned.