cpalockator: Thread-Modular Analysis with Projections

Our submission to SV-COMP’21 is based on the software verification framework "Image missing" and implements the extension to the thread-modular approach. It considers every thread separately, but in a special environment which models thread interactions. The environment is expressed by projections of normal transitions in each thread. A projection contains a description of possible effects over shared data and synchronization primitives, as well as conditions of its application. Adjusting the precision of the projections, one can find a balance between the speed and the precision of the whole analysis.Implementation on the top of the "Image missing" framework allows combining our approach with existing algorithms and analyses. Evaluation on the sv-benchmarks confirms the scalability and soundness of the approach.

Closing the Gap Between Lookahead and Checkpointing to Provide Hybrid Synchronization

Hybrid synchronization provides more in-depth details about real distributed systems. However, several advances in algorithms to provide synchronization between local processes brings new difficulties to integrate into existing simulation architectures. This paper explores an alternative architecture to provide hybrid synchronization. We present optimistic and conservative synchronization primitives and design mechanisms to enable LPs to cooperate during the execution of a simulation. The results show that our primitives improve the simulation in terms of rollback-time and idleness.

Evolving System Bottlenecks in the as a Service Cloud

The web ecosystem is rapidly evolving with changing business and functional models. Cloud platforms are available in a SaaS, PaaS and IaaS model designed around commoditized Linux based servers. 10 billion users will be online and accessing the web and its various content. The industry has seen a convergence around IP based technology. Additionally, Linux based designs allow for a system wide profiling of application characteristics. The customer is an OEM who provides Linux based servers for telecom solutions. The end customer will develop business applications on the server. Customers are interested in a latency profiling mechanism which helps them to understand how the application behaves at run time. The latency profiler is supposed to find the code path which makes an application block on I/O, and other synchronization primitives. This will allow the customer to understand the performance bottleneck and tune the system and application parameters.

Linearizable Wait-Free Iteration Operations in Shared Double-Ended Queues

Shared data object implementations that allow non-blocking concurrent operations are useful for in-memory data-processing, especially when they support consistent bulk operations like iterations. We propose an algorithmic implementation for concurrent iterators on shared double-ended queues (deques), building on and complementing a known lock-free deque implementation by M. Michael. The proposed construction is linearizable and wait-free. Moreover, it is read-only, so it does not execute expensive synchronization primitives and it does not interfere with update operations.