high performance computing on graphics processing units: hgpu.org

Computation outsourcing is an increasingly successful paradigm today. Private and public organizations, as well as common users, can access a large number of economically viable resources to perform the desired computations or access data. The cloud approach allows outsourcers to offer on-demand scalable services to third parties or to perform large computations without high server farm maintenance costs. Scientific computing is one of the beneficiaries of such novel scenarios. Stemming from the first in-house built clusters, distributed computing has flourished over the years also thanks to the Internet. Grids before, and later clouds, have enhanced the storage and computation potentiality. At the same time, to overcome Moore’s law real-world limitations, multicore and manycore hardware has emerged thus allowing parallel computing on a chip. Again, scientific computing in particular has benefited from the speed-up of such novel technology, especially when combined with outsourced cloud computing. As a further opportunity, increasingly powerful mobile devices are now widespread and feature multiple execution cores. They are pervasively immersed in the cloud and can host a large number of applications. These handheld devices can provide additional data acquisition interfaces and pervasive computational resources for heterogeneous workloads. In such a novel scenario it is of primary importance to provide adequate security and privacy guarantees to both providers and users of outsourced services. The problem is complex, as present cloud landscape actors (cloud providers, service providers, and service users) have different requirements and objectives. The first category aims at offering cost-efficient computation while avoiding misuse, the second category aims at reducing costs while at the same time maximizing revenue, the last category aims at reducing the expenses for the accessed services while being guaranteed that the computation is secure, reliable and correct. As a common threat to all these categories lies malicious software, increasingly smart and stealth, affecting a wide range of devices from servers to desktops and even mobile nodes. Such malware can access distributed resources, alter computation outcome and data, leak information. As such, it is of paramount importance to prevent, detect and react to such threats in the smartest possible way. This thesis addresses the security issues involved with computation outsourcing over distributed heterogeneous cloud nodes. It investigates different directions and proposes possible approaches to their solutions in a variety of scenarios. We expect the approaches and achievements presented here to pave the way to further research in the field.