Network architects need to budget for and design their networks for the peak busy hour of utilization, which will grow 5x in the next few years. The peak utilization is growing significantly faster than average utilization, driven by our real-time world that reacts to major events.
Major events can be supply chain disruptions, failure of an assembly line, major weather system, product defect being reported by multiple customers on-line, …. When a major event occurs, people need to see what is going on and collaborate with others to manage the situation. It is at these critical moments that the network must have the capacity to handle real-time and near real-time video, which will consume 95% of network bandwidth.
One example of bandwidth growth is for contact center agents. Currently, enterprise network architects plan on utilizing an average of 200Kbps per agent with 140Kbps for applications delivered via a virtual desktop with 720dpi screen resolution and 60Kbps for Opus based high quality voice. The busy hour is 2.5x busier than average call volumes. By 2020, as contact center agents utilize augmented reality and video based customer interaction at a resolution of 1080dpi, the average bandwidth per agent is predicted to be 640Kbp. The busy hour will be 3.5x busier than average, driven by social media and near real-time events such as a sales promotion of a product/service or a defect.
According to Cisco, busy hour Internet traffic is forecasted to grow 5x by 2020, compared with average utilization which will grow 3x in this time period. 2020 is also the timeframe where 5G cellular will become widely deployed and IoT is forecasted to grow from 6.5B to 30B devices.
Network architects should adopt the following strategies to deal with this explosive growth:
- Overbuild network access – For critical business sites and sites with over 50 people, ensure Ethernet access over fiber with the ability to easily grow from 10Mbps to 10Gbps. Broadband network access works only for smaller and non-critical sites.
- On demand pricing – Change telecom contracts from buying a peak amount of bandwidth to unlimited and bill based on 95th percentile of utilization, which is the model that is used with inter-networking between service providers.
- Do not rely on a single network – Instead of bringing 1-2 networks into an enterprise, enterprises should connect to carrier neutral co-location services and take advantage of the 1,000+ global fiber networks and buy bandwidth at wholesale rates.
- Realize that CAC & QoS have limitations – Dynamic and intelligent management of traffic when congestion occurs instead of doing random early packet drops. Not all critical voice, video, or application sessions are equal and require a segmentation strategy based on security trust and business criticality, not just DSCP classification.
- Improve reporting & analytics – Instead of just monitoring utilization, tracking TCP & UDP packet drops, which is a better indicator of application performance than tracking bits per second.
Too many network architects’ base bandwidth forecasting on average utilization. They then rely on call admission control and quality of service to manage traffic during the peak business hour. While this methodology worked in yesterday’s transaction based networks, future networks are interaction based, which consumes 5x more bandwidth.