QuickStorage: A Write-optimized Efficient Storage System for Trusted Execution Environments

Abstract

With the increasing importance of privacy protection and the growing risks of cloud data leakage and tampering, application and data security have received extensive attention, which is even more necessary for sensitive applications such as identity authentication, multi-party data collaboration, and online financial service system. Trusted Execution Environments (TEEs) can ensure the security of the code and data running in it. Still, the security of the persistent data needs to be guaranteed by the software in the TEE. This paper presents QuickStorage, a storage system that enables sensitive applications running in TEEs to store data safely and efficiently transparently. QuickStorage draws on the idea of the log-structured storage system. Compared with the previous system, the I/O performance of QuickStorage has been dramatically improved through the new design. We implement QuickStorage on the Intel hardware platform and provide two compaction policies, the classic compaction policy and the leveled compaction policy, for log-structured merge-trees (LSMT) in the index area. We conducted intensive tests on systems using two compaction policies to demonstrate the different advantages of compaction policies. The results show that no matter which LSMT compaction policy is adopted by the storage system, write performance is an order of magnitude improvement compared with the previous storage system. Moreover, the overall read performance is also good. The two compaction policies have various advantages and disadvantages in reading and writing, allowing upper-layer applications to choose a more suitable one to meet their various application scenarios.