> Projects > MAP

MAP (Measurement, Analytics, and Profiling)

MAP (Measurement, Analytics, and Profiling) is a usage analytics project aimed at understanding the consumption of mobile and broadband data. MAP provides basic usage statistics and focuses on three directions: capacity utilization, user mobility, and application-specific usage. We've partnered with several ISPs to analyze their data, revealing underlying usage patterns and correlations.

Capacity Utilization

When aggregated across users, demand for capacity on ISP networks generally follows a daily cyclical pattern: in any given hour, the demand on different days is fairly consistent. We can exploit this pattern to predict future demand by using exponential smoothing techniques. Such prediction algorithms, which generally perform with less than 20% error, can aid ISPs in provisioning future capacity on their networks. Furthermore, this high predictability is a useful metric of usage consistency over time.

While demand prediction and consistency are useful metrics, ISPs' operational and capital costs are concentrated at the peak demand over the day. Thus, MAP focuses on the amount of peak traffic from day to day, as well as the distribution of times during which this peak occurs. These consistency measures can help ISPs anticipate the peak demand, as well as devise strategies for lowering the peak. For instance, if usage drops off quickly after the peak, time-dependent pricing strategies may be effective in incentivizing users to shift some demand to off-peak hours, in exchange for some monetary reward.

User Mobility

Our analysis of capacity utilization can be carried out on both an overall network level or at an individual base station level. Since the bottleneck congestion occurs at the level of a few aggregated base stations, this more local examination can yield the most useful information on network congestion. We examine the average usage patterns at different base stations and group them into different clusters, revealing similarities in usage. For instance, some base stations, e.g., those near business districts, may be heavily used by business users, while others cater more towards recreational users. MAP can utilize the base station usage patterns to help distinguish the behavior of these demographics, allowing ISPs to develop services targeted towards relieving congestion on specific base stations.

In addition to its data analysis, MAP provides a visualization of the real-time congestion conditions at different base stations. ISPs can view the different base stations on a map, color-coded by congestion level, and select specific ones to view their particular usage statistics. These statistics may include application-specific usage (in some countries, ISPs are prohibited from accessing this information).

Application-Specific Usage

Certain applications drive network congestion more than others—video, for instance, consumes more data than Internet browsing. Yet the exact composition of traffic may vary across different ISPs and different areas of the network (e.g. business versus suburban areas). At its most basic level, MAP can provide these usage distributions. We also examine the number of active users for each application type over time—for instance, does high video streaming usage arise from multiple people watching short videos in the same hour, or from a few people with sustained video streaming? ISPs can use this data when targeting certain applications for congestion alleviation: it is easier to target a few heavy users of particular apps than to change several users' behavior.

MAP goes beyond simple distributions to investigate correlations in application usage—for instance, does heavy video consumption correlate with peer-to-peer downloads? ISPs looking to offer application-specific pricing can use these correlations to target particular apps. For example, if P2P downloads and video streaming are highly correlated, ISPs could offer a data plan that discounts them both. These data plans could in turn be targeted to specific users—MAP's clustering algorithms can group users by application-specific usage. Certain groups can then be offered tailored data plans to accommodate their usage behavior and influence them to reduce peak network congestion.