RepComp - Replicated Components for Improved Performance or Reliability in Multicore Systems
Jan 2010 - Dec 2012
Multi-core processors have become widely available in the last couple of years. Although these processors are becoming increasingly powerful, with the integration of an increasing number of cores, it also makes it harder for applications to exploit this increasing power. Unlike before, that any application would automatically benefit from the increase of clock speed, with multicore CPUs, programs must include multiple concurrent threads of activity to take benefit from the multiple cores available.
In this project we intend to explore the idea of diverse replication of software program components to achieve two goals.
First, we expect to be able to improve the overall performance of an application by relying on the result obtained from the fastest replica in each operation (assuming that for different operations of a software component, the fastest component is not always the same).
Second, we expect to be able to provide fault-tolerance to buggy implementations by using Btzantine fault-tolerance replication techniques.
Multi-core processors have become widely available in the last couple of years. Although these processors are becoming increasingly powerful, with the integration of an increasing number of cores, it also makes it harder for applications to exploit this increasing power. Unlike before, that any application would automatically benefit from the increase of clock speed, with multicore CPUs, programs must include multiple concurrent threads of activity to take benefit from the multiple cores available.
In this project we intend to explore the idea of diverse replication of software program components to achieve two goals.
First, we expect to be able to improve the overall performance of an application by relying on the result obtained from the fastest replica in each operation (assuming that for different operations of a software component, the fastest component is not always the same).
Second, we expect to be able to provide fault-tolerance to buggy implementations by using Btzantine fault-tolerance replication techniques.
