Given the high demand of computational resources by Web IR Systems, many-core processors can be a promising technology nowadays.
Novel many-core processors are programmable accelerators classified as high-parallel sharedmemory architectures capable of high performance. Such architectures born to solve specific problems, such as the graphical ones. Their exploitation to solve a wider range of problems, in general, requires to design the related algorithms for such architectures. The design of efficient algorithms for such architectures is also made more difficult by the lack of computational models that adequately abstract their features.
Models that do not give a realistic abstraction can lead to unrealistic expectations for the algorithm designed, and thus to evaluate theoretically optimal algorithms that are not optimal on any realistic machine.
In this PhD thesis we aim to study the potentiality of many-core architectures to realize large-scale distributed Web IR systems. The challenge of effectively and efficiently programming many-core processors can be addressed exploiting a computational model that captures their essential features.