| dc.description.abstract | Internet of Things (IoT) has allowed embedded devices to connect to the vast internet network worldwide. The amount of data produced and exchanged between them is growing exponentially and with the present hardware and software architecture it is difficult to support them. With billions of IoT devices waiting to be connected in the near future, it is necessary to build infrastructure for the upcoming change as the energy and cost associated with the continuous transmission, classification and storage will be huge. We need to build an efficient framework that can scale easily, follow consistent protocol, maintain security and save resources.
The thesis focuses in solving the major upcoming problems of the Internet of Things by proposing a lightweight framework which resides in both the server and the end device as server client model. The framework has the following benefits – it reduces network congestion, reduces data consumption and maintains security. The framework resides on the data and communication layer, classifying the data into known patterns - Motifs. We have used modified Hidden Markov Model to classify the sensor data into Motifs. The framework transfers only the motifs attributes information instead of complete sensor data. Thus the data can now be compressed by orders of magnitude into these classes of recurrent patterns. It not only saves on data storage but also on network transmission. It helps us to create a state based model and in anomaly detection and security.
We also optimize Partial Homomorphic Encryption based on El-Gamal Algorithm using OpenCL, OpenMP, SIMD, batch processing, Karatsuba algorithm and used to secure the framework while allowing simple computation to be performed on the encrypted data. | en |
