Machine Learning-based Geolocation Of Cloud Data

After years of development,cloud storage has increasingly played an essential role of Internet life.A large number of individuals and business users have outsourced their data to the cloud for the purpose of reducing local storage maintenance costs while accessing their data more conveniently.However,in spite of benefiting from cloud storage,the users lose the direct control over their data,raising a series of security issues,among which the storage locations of data are especially vital under certain conditions.National citizen information and healthcare data contain a lot of privacy and even involve national security,so their storage locations must meet certain restrictions.At present,some countries have formulated relevant laws and regulations to explicitly limit the storage locations of sensitive data.However,the cloud service provider may store the data in some location with lower maintenance cost for economic benefit,which violates the user's requirements for the storage locations.Therefore,there is a urgent need for reliable techniques to geolocate data stored in the cloud.However,only a few research results related to this field have been published,and existing schemes are deployed and experimented in North America or Europe rather than China.In this thesis,according to the actual scene of cloud storage,we design and implement a geolocation of cloud data scheme.Firstly,we describe the participators,establish a system model,introduce the features related to locations,analyze the potential attacks,and propose the security goals of cloud data geolocation.Then we design a machine learning-based geolocation of cloud data scheme,which combines the principles of IP geolocation techniques with data integrity verification and uses machine learning to train the relationship between features and locations.In the construction stage of the proposed scheme,we construct two implementations,which have differences in computation and communication overhead.In addition,we analyze the security of the proposed scheme under the potential attacks.Next,the proposed scheme is deployed in China's environment,we collect the experiment data,and test the feasibility of the scheme by experiments.Because of the popularity of the solid state drive(SSD)in the cloud,we conduct experiments with the traditional hard disk drive(HDD)and SSD respectively in the simulated cloud environment,analyzing the influences of different hardwares,related parameters and attacks on the geolocating accuracy of the two implementations of the scheme.And we also conduct experiments in the real cloud environment,analyzing the geolocation accuracy and security of the two implementations in two circumstance where attacks exist or not.Finally,we compare the two implementations of the proposed scheme,and compare them with the existing schemes in aspects of various performance.The results of experiments show that our proposed machine learning-based scheme achieves the city-level geolocation of cloud data and meets the proposed security goals.The geolocation accuracy of the scheme in condition of SSD is higher than that in condition of HDD.And our scheme has higher accuracy than the state-of-the-art schemes.