E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Current Record: 75 of 824

[Add to Basket]

Identifier

000443695

Title

Atlas: automated scale-out of trust-oblivious systems to trusted execution environments

Alternative Title

ΑΤΛΑΣ: αυτοματοποιημένη κλιμάκωση μη-ασφαλών συστημάτων σε αξιόπιστα περιβάλλοντα εκτέλεσης

Author

Αναγνώπουλος, Γεώργιος Α.

Thesis advisor

Πρατικάκης, Πολύβιος

Reviewer

Ιωαννίδης, Σωτήριος
Μαγκούτης, Κωνσταντίνος

Abstract

Trusted Execution Environments (TEEs) offer important security benefits to applications that combine on- and off-premise components. Acquiring these benefits, however, requires significant developer effort in order to identify and port security sensitive application components inside a TEE. The lack of high-level TEE APIs enforces the community to utilize low-level interfaces, commonly offered in C/C++, dealing with the complexities that low-level programming languages come with; i.e. memory handling, debugging and maintenance. Also, these security benefits come with a performance trade-off due to the added encryption/decryption schemes, integrity checking and protected memory limitations. This work presents Atlas, a system for automatically scaling out components on TEEs, using a high-level programming language, namely JavaScript. Our system uses program transformations to offload the sensitive function calls of a given application and distribute the load among trusted nodes. This is achieved by embedding JavaScript’s run-time environment within the TEE and performing the appropriate optimization in order to achieve remote function execution. We evaluate Atlas using a set of language-specific algorithms and cryptographic suites as well as with three real-world applications written in JavaScript. This results show that Atlas is able to scale-out legacy applications, originally not developed with TEE capabilities, with significant performance benefits. Our system is able to perform the execution up to 7 times faster compared to the vanilla QuickJS JavaScript interpreter, using ten TEE-enabled remote nodes, while also providing elasticity characteristics, all achieved with minimal developer effort.

Language

English

Subject

Cloud

Intel

Javascript

QuickJS

SGX

Security

TEE

Trusted

Ασφάλεια