Communication and application servers are usually separated from database servers. Each and every message requires communication over the network causing considerable traffic between the servers and databases. In high-load scenarios, this can result in considerable network traffic and can become a bottleneck. In the best case this can be solved by optimising network settings, and when optimised by installing more network interface cards and/or increasing network bandwidth. Expensive.
With more and more smart devices (smart phones, tablets) connecting to cellular and mobile networks (GSM, LTE, etc) networks, an exponential growth in data traffic can be expected from that too. Shannon’s noisy channel coding theorem will hold true, causing another bottleneck against this rise in data demand. Allocating more spectra to meet data demands will not be a viable solution, and is not under the control of the owners/maintainers of a private network anyway. Because WiFi networks have low cost and are everywhere and the WiFi spectrum is not licensed and is re-usable, the current solution is often to offload, on-the-fly, some or all of the data traffic to the WiFi network (almost all of the smart devices come with WiFi support), potentially causing another bottleneck against this rise in data demand, now on WiFi networks, and with some problems (untrusted access points, authentication, and addressing).