An NVX framework that combines selective binary rewriting with a novel event-streaming architecture to significantly reduce performance overhead and scale well with the number of versions.

Overview

With the widespread availability of multi-core processors, running multiple diversified variants or several different versions of an application in parallel is becoming a viable approach for increasing the reliability and security of software systems. The key component of such N-version execution (NVX) systems is a runtime monitor that enables the execution of multiple versions in parallel.

Unfortunately, existing monitors impose either a large performance overhead and/or rely on intrusive kernel-level changes. Moreover, none of the existing solutions scales well with the number of versions, since the runtime monitor acts as a performance bottleneck.

Varan is an NVX framework that combines selective binary rewriting with a novel event-streaming architecture to significantly reduce performance overhead and scale well with the number of versions, without relying on intrusive kernel modifications.

Our evaluation shows that Varan can run NVX systems based on popular C10k network servers with only a modest performance overhead, and can be effectively used to increase software reliability using techniques such as transparent failover, live sanitization and multi-version execution.

Research Support

This research project is generously sponsored by Google through a PhD Fellowship.

Publications

  • Towards Deployment-Time Dynamic Analysis of Server Applications

    Luís Pina, Cristian Cadar

    Proceedings of the 13th International Workshop on Dynamic Analysis (WODA 2015)

  • Varan the Unbelievable: An Efficient N-version Execution Framework

    Petr Hosek, Cristian Cadar

    20th International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'15)