Apurv Mehra bio photo

Apurv Mehra


IIITD Gmail Twitter LinkedIn Github


Mobile CrowdSensing (MCS) applications rely on physical sensor data collected by mobile devices of a large number of participants. Participating devices are required to execute sensing tasks with certain requirements such as geographical location or data quality. Often, it is sufficient to execute the tasks on a subset of devices to avoid data redundancy and improve energy efficiency. Since mobile devices are resource-constrained, it becomes important to ensure a load-balanced allocation of sensing tasks to this subset of participants, which will eventually ensure longer lifetimes of devices and therefore, higher task service rates. To take into account the resource-constrained nature of participants, we consider the dynamic parameters of participants (such as battery level and link speed) and their static parameters (such as sensor power rates and the number of sensors available) to propose a load-balanced task allocation algorithm for MCS systems. Our algorithm is based on the model of device ‘richness’ which is a measure of their available parameter levels. Based on the richness, we propose two variants of our algorithm - greedy allocation, and weighted-random allocation. We compare our algorithms with three baseline approaches using simulations. Using a quantitative measure of inequality distribution - Gini coefficient - we show that our algorithm achieves perfect load balancing when the participants have homogeneous configurations. Further, using our approach, we are able to improve the lifetime of the system by approx. X% which eventually results in a Y% improvement in the task service rate.


  1. We propose a model to evaluate the richness of devices in terms of their capabilities (resource levels). Our model is inspired by the Linux-based Completely Fair Scheduler (CFS) which ensures fair allocation of CPU time to processes based on their priorities.
  2. We propose two variants, NV-Greedy and NV-Weighted, of the load-balanced task allocation algorithm based on our device richness model and compare their performance with three baseline algorithms - Random, RoundRobin, and MinAST , using extensive simulations.
  3. We objectively evaluate the effectiveness of our algorithm using a quantitative measure - Gini coefficient . Gini coefficient is commonly used in the field of economics to measure the inequalities in income distribution of nature. We use it to measure the inequalities in the distribution of richness levels of devices.

System Design

Mew Architecture


Our results show that the proposed variants improve the fairness in task allocation by upto 2-3%, which results in an additional availability of upto 665 minutes of devices with the system while improving the location accuracy of collected sensor data. This is a significant gain which will increase with the increase in number of devices available with the crowdsensing system.


  1. Achieving Load Balanced, Dynamic Task Allocation in Mobile Crowdsensing Systems.pdf