I found the following explanation of Sprint's Network Management Practices very helpful. Starting Deprioritization at 50Gb is great.


How does Sprint manage its network?
Sprint employs a holistic approach to managing congestion on its data network. Sprint’s first goal is to avoid congestion altogether by directing traffic to the best available spectrum resources and cell sites. Sprint also attempts to avoid congestion by managing tonnage on its network. Finally, when congestion does occur, meaning that the demand on a particular sector temporarily exceeds the ability of that sector to meet the demand, Sprint relies on the radio scheduling software provided by Sprint’s hardware vendors to allocate resources to users.

Techniques to Direct Traffic to the Best Available Spectrum Resources and Cell Sites:
All mobile networks, including Sprint’s, employ a Radio Access Network (“RAN”) that manages connectivity between mobile client devices and the core network. The RAN functions to identify mobile devices permitted to access the network and their locations and assigns the mobile device to an available frequency band and cell site serving the location. The RAN also controls device “hand off” between neighboring cell site resources to balance load across network resources or as a mobile device moves from one location to another. Sprint’s RAN manages connections between mobile devices and cell sites operating on multiple frequency bands (800 MHz, 1.9 GHz, 2.5 GHz) and multiple air interfaces (CDMA, EVDO, LTE). As part of managing those connections, Sprint’s RAN is designed to dynamically connect customers to the best available spectrum resources and cell sites—and reassign those connections as circumstances change.

Managing Tonnage:
Some of Sprint’s traffic management efforts are aimed at avoiding congestion by managing the total volume of data transmitted, referred to colloquially as “tonnage.” Sprint may work with high-volume content providers to help ensure that their content is delivered in a way that uses Sprint’s network in an efficient manner. In addition, some applications may reduce or increase their traffic volume depending on network conditions. For example, some streaming video applications employ adaptive bitrate protocol to stream video. These applications automatically and continuously monitor the available bandwidth and adjust the streaming video bitrate to current user conditions. Depending on available bandwidth, users may notice differences in video streaming quality as the application adjusts the video streaming bitrate to account for changing channel conditions.

Allocating Resources During Times of Congestion:
Despite our best efforts to prevent congestion through managing tonnage and directing customers to the best available network resources, the demand on a particular network sector sometimes temporarily exceeds the ability of that sector to meet the demand. During these times, Sprint relies on the radio scheduling software provided by Sprint’s hardware vendors to allocate resources to users. This radio scheduling software includes a set of generic fairness algorithms that allocate resources based on signal quality, number of users, and other metrics. These algorithms are active at all times, whether or not the cell is congested; however, during times of congestion, the algorithms operate with the goal of ensuring that no single user is deprived of access to the network.

Quality of Service (QoS):
To help protect against the possibility that unlimited data plan customers that use high volumes of data may occupy an unreasonable share of network resources, Sprint employs network prioritization or QoS on the Sprint network. Customers who choose unlimited data handset plans launched on or after October 16, 2015, or customers who choose to upgrade their handsets or activate new lines of service on or after October 16, 2015, and are on unlimited data plans, that use more than 50GB (35GB for Virgin Mobile and Boost Mobile) of data during a single billing cycle (the QoS Threshold) will be de-prioritized for the remainder of that billing cycle as compared to other customers at times and places where the availability of network resources is constrained. Affected unlimited data customers will continue to be able to enjoy unlimited amounts of data without the worry of overage charges or hard, full time bandwidth reductions. Customers subject to de-prioritization may experience reduced throughput and increased latency compared to other customers on the constrained site and as compared to their normal experience on the Sprint network. Unlimited customers may also notice temporary changes in the performance of data intensive applications such as streaming video or online gaming when subject to de-prioritization. These temporary reductions in performance will only occur at times and places where capacity is constrained. Performance will return to normal as soon as the resource constraints have been relieved or the customer has relocated to a non-constrained location. Unlimited data customers potentially subject to lower QoS will be notified when their individual data usage reaches approximately 75% of the QoS Threshold so that they may modify their usage to avoid network management practices that may result in slower data speeds. We will also notify customers when they have reached the QoS Threshold and are now subject to de-prioritization.

Real-Time Traffic Management System (RTTMS):
Another tool Sprint has deployed to help improve the customer experience for the majority of customers during times and places where network resources are constrained is Sprint’s RTTMS. When a Sprint site is resource constrained, RTTMS works to identify the users consuming the most site resources, and reallocates a portion of the resources serving the data intensive users to other users engaged in less resource intensive activities. When RTTMS is activated on a resource constrained site, customers engaged in data intensive applications may experience reduced performance until the site becomes unconstrained or the user moves to a non-resource constrained site. Users of the resource-constrained site that are not engaged in data intensive actives should expect a better experience than they would have on the site if RTTMS not activated. Sprint’s RTTMS is user, content, application, and device agnostic and is intended to provide an improved experience for the majority of users when resources are constrained.