24.
02.
2014
ELK Herder
Florian Pester
TU Dresden
Verteidigung der Diplomarbeit
Today's software is prone to faults in the underlying hardware. We need
fast and flexible solutions that deal with these faults in order to keep
software reliable, even on commercial-off-the-shelf systems. Replication
in virtual machines with a minimal proxy operating system provides for
flexibility, because it lets users choose, which applications to
replicate. Modern COTS hardware provides multiple CPU cores and
virtualization support, minimizing replication overhead. I implemented
ELKVM, a library that runs applications in a virtual machine with 2.2%
runtime overhead. I present ELK Herder a replication manager, which runs
these applications in a triple-modular-redundant fashion, providing
fault tolerance and recovery, with 16.3% total runtime overhead. These
developments make fault tolerance possible and affordable on low-end
servers, and most consumer devices.
|