Secure Cloud Computing

Research Mentors: Marten van Dijk and John Chandy

Cloud computing entails a sacrifice of control; its users give up configuration and management oversight of the infrastructure that contains their data and computing resources. To make cloud computing trustwor- thy we need to find system solutions that allow users to verify that their outsourced private data remains private (even if untrusted third party programs are allowed to compute on the data), can be recovered from failures and attacks, and is indeed stored at a user-specified geographical location (which is often required for legal reasons). By assuming a small trusted computing base in the form of minimal trusted HW within data centers, is it possible to design an efficient user interface that is capable of verifying these necessary guarantees?

Components for Student Participation

REU students will participate in the following projects. Power Efficient Oblivious RAM. Users generally come in two forms: enterprises who outsource e.g. their data base computations (on financial information, medical records, etc.) and individuals who wish to store their private data (tax forms, private pictures, documents, etc.) at some third party server (e.g. Dropbox). From the individual user’s perspective, a user SW interface/app should fit his smartphone in that it operates efficiently and is, in particular, power efficient. Besides managing a decryption key and checking integrity and freshness, the interface should also ideally eliminate privacy leakage through the user’s access patterns to his outsourced data. In our recent work we developed Path-ORAM which is a simple and practical system that exactly provides this guarantee. To adopt Path-ORAM in a smartphone setting, we need tricks to reduce its power consumption due to the large number of needed AES decryptions (e.g. dummy blocks do not need to be decrypted if the interface knows their location and Path-ORAM write-operations can happen less frequently if the interface keeps severable MB local state).

REU students will research, simulate and implement a Path-ORAM interface that best suits the smartphone setting. Through this process students learn the most recent advances in security research as applied to cloud computing and storage. Proof of Geolocation. One of the main problems in data management is proving (for legal reasons) that an outsourced computation uses data that is stored locally at the data center in a user-specified geographical location (e.g., enterprises may only outsource storage of salary and other information of their employees if this information is stored within the appropriate legal jurisdiction). We want to design a stronger solution that allows an enterprise/user to remotely verify that sufficient information about his data is stored at a specific (in terms of geolocation) data center such that all the user’s data can be reconstructed from this information. We assume at least one hardware root of trust in the data center (a server with TPM/TXT), which the user exploits by interacting through a protocol that randomly fetches data samples from the data center’s disks.

The root of trust measures the time taken by the protocol; if below a certain threshold, the user may conclude (under appropriate rational adversarial assumptions) that the protocol could not have finished if a significant part of his data would have been stored non-local to the data center. If the user’s data is stored with sufficient redundancy, then the user knows that all his data can be reconstructed from information that is stored within the data center alone. Through this project REU students will learn about TPM/TXT technology and data center infrastructure, and design and experiment/simulate a small proof of concept.