Search

Cloud technology is an attractive infrastructure solution that provides customers with an almost unlimited on-demand computational capacity using a pay-per-use approach, and allows data centers to increase their energy and economic savings by adopting a virtualized resource sharing model. However, resources such as graphics processing units (GPUs), have not been fully adapted to this model. Although, general-purpose computing on graphics processing units (GPGPU) is becoming more and more popular, cloud providers lack of flexibility to manage accelerators, because of the extended use of peripheral component interconnect (PCI) passthrough techniques to attach GPUs to virtual machines (VMs). For this reason, we design, develop, and evaluate a service that provides a complete management of cloudified GPUs (cGPUs) in public cloud platforms. Our solution enables an effective, anonymous, and transparent access from VMs to cGPUs that are previously scheduled and assigned by a full resource manager, taking into account new GPU selection policies and new working modes based on the locality of the physical accelerators and the exclusivity when accessing them. This easy-to-adopt tool improves the resource availability through different cGPUs configurations for end-users, whilst cloud providers are able to achieve a better utilization of their infrastructures and offer more competitive services. Scalability results in a real cloud environment demonstrate that our solution introduces a virtually null overhead in the deployment of VMs. Besides, performance experiments reveal that GPU-enabled clusters based on cloud infrastructures can benefit from our proposal not only exploiting better the accelerators, but also serving more jobs requests per unit of time.

Nowadays, video coding and transcoding have a great interest and important impact in areas such as high-definition video and entertainment, healthcare and elderly care, high-resolution video surveillance, self-driving cars, or e-learning. This growing demand for high-resolution video boosts the proposal of new codecs and the development of their encoders that require high computational requirements. Therefore, new strategies are needed to accelerate them. Cloud infrastructures offer interesting features for video coding, such as on-demand resource allocation, multitenancy, elasticity, and resiliency. This paper proposes a cloud-based distributed architecture, where the network and the storage layers have been tuned, to accelerate video encoders over an elastic number of worker encoder nodes. Moreover, an application is developed and executed in the proposed architecture to allow the creation of encoding jobs, their dynamic assignment, their execution in the worker encoder nodes, and the reprogramming of the failed ones. To validate the proposed architecture, the parallel execution of existing video encoders, x265 for H.265/HEVC and libvpx-vp9 for VP9, has been evaluated in terms of scalability, workload, and job distribution, varying the number of encoder nodes. The quality of the encoded videos has been analyzed for different bit rates and number of frames per job using the Peak Signal-to-Noise Ratio (PSNR). Results show that our proposal maintains video quality compared with the sequential encoding while improving encoding time, which can decrease near 90%, depending on the codec and the number of encoder nodes.

Radio frequency signals are commonly used in the development of indoor localization systems. The infrastructure of these systems includes some beacons placed at known positions that exchange radio packets with users to be located. When the system is implemented using wireless sensor networks, the wireless transceivers integrated in the network motes are usually based on the IEEE 802.15.4 standard. But, the CSMA-CA, which is the basis for the medium access protocols in this category of communication systems, is not suitable when several users want to exchange bursts of radio packets with the same beacon to acquire the radio signal strength indicator (RSSI) values needed in the location process. Therefore, new protocols are necessary to avoid the packet collisions that appear when multiple users try to communicate with the same beacons. On the other hand, the RSSI sampling process should be carried out very quickly because some systems cannot tolerate a large delay in the location process. This is even more important when the RSSI sampling process includes measures with different signal power levels or frequency channels. The principal objective of this work is to speed up the RSSI sampling process in indoor localization systems. To achieve this objective, the main contribution is the proposal of a new MAC protocol that eliminates the medium access contention periods and decreases the number of packet collisions to accelerate the RSSI collection process. Moreover, the protocol increases the overall network throughput taking advantage of the frequency channel diversity. The presented results show the suitability of this protocol for reducing the RSSI gathering delay and increasing the network throughput in simulated and real environments.

Indoor RF-based localization using fingerprint mapping requires an initial training step, which represents a time consuming process. This location methodology needs a database conformed with RSSI (Radio Signal Strength Indicator) measures from the communication transceivers taken at specific locations within the localization area. But, the real world localization environment is dynamic and it is necessary to rebuild the fingerprint database when some environmental changes are made. This paper explores the use of different interpolation functions to complete the fingerprint mapping needed to achieve the sought accuracy, thereby reducing the effort in the training step. Also, different distributions of test maps and reference points have been evaluated, showing the validity of this proposal and necessary trade-offs. Results reported show that the same or similar localization accuracy can be achieved even when only 50% of the initial fingerprint reference points are taken.