Spectrum Aware Cloud Offloading
PI: K.P. Subbalakshmi, Prof. Dept. of E.C.E, Stevens Institute of Technology
Co-PI: R.N. Uma, Assoc. Prof. Dept. of Mathematics and Computer Science, North Carolina Central University
Syed Eman Mahmoodi, PhD Candidate, Stevens Institute of Technology
The Spectrum Aware Cloud Project will address several issues arising in mobile computing in the 5G world. NSF's CloudLab resources will be used to extensively test these solutions.
Fig. 1: Scheduling Model for Cloud Computing in a Multi-Component Mobile Application
A. Single Radio Offloading
First we consider the single radio case, with applications that can be decomposed into components with general dependencies (as opposed to simple chain dependencies) between components as shown in Figure 1. The goal is to process the application within the specified amount of time while minimizing the energy expenditure of the mobile device. In such problems, an effective trade-off must be struck between the amount of energy spent in processing and data transfer against the allotted time limit to fully process the application.
We formulate the problem as a 0-1 integer linear program using time-indexed variables xpjt that is assigned the value 1 when component j is processed in time period [t, t+1) on either the cloud (p = 1) or the mobile device (p = 0). Such a formulation affords us the scope to easily expand the problem in our future studies to consider additional constraints as dynamic spectrum allocation in multi-radio enabled communications. Additionally, solutions to the 0-1 integer linear program are guaranteed to satisfy all the constraints of the problem as well as serve as a guide to designing time-efficient heuristics.
The cloud services provided by NSF's CloudLab will be used for two purposes: to solve the 0-1 integer program as well as to emulate the execution of the application by the mobile device to calculate the actual energy expense and time expense. To thoroughly investigate our techniques and heuristics and to study their scalability, we seek to test our techniques on significantly larger applications.
In particular, we propose to use CloudLab to carry out the following experiments:
- Evaluate our techniques and heuristics on applications with general precedence constraints vs. series-parallel constraints. As it is well known, the underlying scheduling problems for the general-precedence constraint case is conjectured to be as hard to approximate as the vertex cover problem.
- Investigate scalability by considering larger applications - vary number of components in the applications from 100 to 1000. This will also increase the size of the underlying integer programming formulation significantly.
Fig. 2. Spectrum Aware Cloud Computing Infrastructure
B. Multi-Radio Offloading
In the second stage we will advance to scenarios where more than one radio network is available to the mobile user as shown in Fig. 2. Experiments in this case will be broken down into stages with increasing complexity.
- Stage 1: In this first case, we consider an application with simple dependencies (chain rather than graph) between the components. The problem here is to optimally decide which components of an application to offload and which to execute locally, while simultaneously optimizing the percentage of data (associated with this offloading) to be sent via each radio interface. This computation offloading problem is set up as a joint optimization to minimize the energy consumed on the device while at the same time maximizing the radio resources available to the device, under some operational constraints. The will be used to experiment on complex applications with several components.
- Stage 2: In this second scenario, we will increase the complexity of the problem to add the optimization of the downlink transfers as well. Experiments on the and Chameleon will now involve running larger apps with more “expensive” optimization algorithms.
- Stage 3: Putting the pieces together. In this phase, we will include apps with more complex inter-component relationships (similar to the ones studied in II-A) and add multiple radio interfaces to the scenario. Both uplink and downlink optimizations will be considered.
Associated Publications and Patents
Syed Eman Mahmoodi, K.P. Subbalakshmi and Vidya Sagar, "Cloud Offloading for Multi-Radio Enabled Mobile Devices", IEEE International Conference on Communication, June 2015.
Syed Eman Mahmoodi and K.P. Subbalakshmi, “Cognitive cloud off-loader: Real-time method for joint scheduling offloading computation in multi-RAT enabled mobile devices”, Provisional Patent Filed, December 2015.
Syed Eman Mahmoodi, R.N. Uma and K.P. Subbalakshmi, "Optimal Joint Scheduling and Cloud Offloading for Mobile Applications", IEEE Transactions on Cloud Computing, April 2016, Online ISSN 2168-7161, Download.
- Syed Eman Mahmoodi and K.P. Subbalakshmi, "A Time-Adaptive Heuristic for Cognitive Cloud Offloading in Mulit-RAT Enabled Wireless Devices", IEEE Transactions on Cognitive Communications and Networking, Vol. 2, Issue 2, pp.:194-207, June 2016. Download
- Syed Eman Mahmoodi, K.P. Subbalakshmi and R. N. Uma, "Harnessing Spectrum Awareness to Enhance Mobile Computing", ACM 22nd International Coneference on Mobile Computing and Networking (Mobicom) 2016. pp: 460-461. Download
- Syed Eman Mahmoodi and K.P. Subbalakshmi and R.N. Uma, "Spectrum Aware Mobile Computing using Cognitive Networks, Handbook of Cognitive Radio -- Section: Dynamic Spectrum Access and Sharing, Springer, Eds: Dusit Niyato and Ping Wang, EiC: Wei Zhang [invited]
- Syed Eman Mahmoodi, K.P. Subbalakshmi and R. N. Uma, "Spectrum-Awareness in Mobile Computing - Convergence of Cloud Computing and Cognitive Networking", Springer International Publishing.