ABSTRACT
The fact that concurrency bugs are only triggered under certain instruction interleavings makes this class of bugs particularly difficult to find and reproduce. In practice, developers often rely on stress tests, which consist in repeatedly running test cases, hoping to explore different interleavings and to eventually execute the interleavings that trigger the concurrency bugs.
In this talk I will describe the design of a virtual machine monitor, SKI, that allows developers to systematically test operating system kernels for concurrency bugs. In contrast with stress tests, our tool takes full control over the interleavings of the kernel threads and is able to explore the interleaving space systematically, according to its scheduling algorithm. The design of SKI does not require modifications to the tested kernel and our experience using the tool shows that SKI is effective at finding and reproducing real-world kernel concurrency bugs.
SHORT BIO
Pedro Fonseca is a final year graduate student at MPI-SWS and he is broadly interested in improving the reliability of software. Recently he has been working on techniques to help developers find concurrency bugs in complex software. In the past, he has worked in several other areas, including Byzantine fault tolerant systems, accountability, peer-to-peer systems and OS architectures.
More info: http://citi.di.fct.unl.pt/seminar/seminar.php?id=236