Section: New Results

Locality for Application Using Locks on Clusters of Multicore Platforms

The aim of this post-doc work is to study the locality for applications based on read-write locks on clusters of multi-core platforms. We focused on the implementation of the video tracking application  [25] using the Ordered Read Write Locks (ORWL)  [20] model of programming on multi-cores architecture. For several uses, such as, human-computer interaction, security or traffic control, the tracking application aim to locate multiple moving objects over time using a camera. Its processing can be a time consuming process due to the amount of data that is contained in high definition video which leads to decrease the throughput. To overcome this problem it is possible to speed up the processing by exploiting task parallelism of ORWL model. Indeed, the model proposes abstractions of the decomposition in parallel parts (tasks), the synchronization of and the communication between threads. However, we noted some problems which decrease the parallelism scaling thus we introduced different optimizations: stream multiplexing, multiple buffering, etc. We are now working on parallelizing long-running tasks.