Search

We study three models of job scheduling in a distributed server system. For each of them we suggest scheduling algorithms that are computationally efficient and provably achieve a performance objective related to the model. For this we consider jobs to be an abstraction of executable programs that request specific resources e.g. memory, CPU. Resources need to be reserved in one of the servers for the duration the programs run, which is unknown. The first model considers queue-based scheduling algorithms, in which jobs belong to a finite set of types and each type has a separate queue. The scheduling objective under this formulation is to keep the size of all queues bounded, which translates to bounded queuing delay. The two families of algorithms for this model can achieve the objective for the maximum theoretical workload. Most importantly they follow vastly different paradigms and are both viable alternatives depending on what other trade-offs the scheduling has to achieve. The second model considers that resource requirements of jobs come from an unknown distribution. Jobs are queued and the objective is again to keep the number of jobs in the queue bounded and consequently the queuing delay. In this harder formulation there is no previous characterization of the maximum workload that can achieve the objective. We provide such a characterization and algorithms that achieve at least 2/3 of that maximum. Lastly, we consider a model without queues in which jobs are admitted or rejected on arrival, with the goal to maximize the total utility of the jobs that run. Algorithms of this model were proven to achieve at least 1/2 of the maximum, but further analysis suggests that this limit can be as high as 1 − e⁻¹. In all models we made simplifying assumptions that allow us to prove the desired properties of the system, but despite the theoretical nature of this work, we also discuss how the algorithms can be applied and tailored to the needs of different cloud applications. We hope they will eventually inspire improvements to existing cloud infrastructure management deployments.

The DrosophilaRetina has been extensively characterized in the literature in terms of structure,connectivity and function.We present here a detailed description of the algorithms required for afull-scaleparallel emulation of the Retinaincluding (i) the mappingof the visual field onto the photoreceptors, (ii) thephototransductionprocess, and (iii) the parallel processing of the visual field by theentire Retina.We also provide detailed algorithms, theirimplementation and their visual evaluation.

This deliverable updates the scenarios and requirements described in D2.1.2 to include on-demand features identified during connectivity phase development, the usage by driving experiments and experiments funded through the open calls. The document provides a description of the functionality that needs to be developed within the expansion phase, to be included in the V2.0 release.

This deliverable provides an early first iteration on the Scenarios informing the EXPERIMEDIA project, and the Requirements we derive from them. The document provides the EXPERIMEDIA developers will an initial set of functionalities expected to be supported by the facility during the connectivity phase, to be included in the V1.0 release. The document further develops the needs of the three driving scenarios of (live and augmented, live and 3D, and live and online) along with technologies and working practices necessary to support the experimental lifecycle. The document considers the implications of the initial results of methodology D2.1.1 and provides initial requirements for security and data protection from WP.

Summarization: We consider the fully automated behavior understanding through visual cues in industrial environments. In contrast to most existing work, which relies on domain knowledge to construct complex handcrafted features from inputs, we exploit a Convolutional Neural Network (CNN), which is a type of deep model and can act directly on the raw inputs, to automate the process of feature construction. Although such models are limited to handle still 2D inputs, in this paper we appropriately transform video input to incorporate temporal information into each frame. This way our model hierarchically constructs features from both spatial and temporal dimensions. We apply our model in real-world environment, on data taken from Nissan factory, and it achieves superior performance without relying on handcrafted features. Παρουσιάστηκε στο: 23rd IEEE International Conference on Image Processing

Διπλωματική Εργασία που υποβλήθηκε στην σχολή ΗΜΜΥ του Πολ. Κρήτης για την απόκτηση διπλώματος μηχανικού. Περίληψη: Η παρούσα διπλωματική εργασία έχει σκοπό την περιγραφή και χρήση του λογισμικού που αναπτύχθηκε για την ανάλυση δυναμικής ευστάθειας συστημάτων ηλεκτρικής ενέργειας. Το εργαλείο αυτό αναπτύχθηκε κατά το ακαδημαϊκό έτος 2013-2014 από τον Κωνσταντίνο Ραφαήλ Κουσουλίδη, απόφοιτο της Σχολής Μηχανικών Παραγωγής και Διοίκησης του Πολυτεχνείου Κρήτης υπό την επίβλεψη του κ. Φώτιου Κανέλλου, για την ανάλυση ροής φορτίου σε συστήματα ηλεκτρικής ενέργειας. Σε συνέχεια της εργασίας αυτής, αναπτύχθηκε το τμήμα της ανάλυσης δυναμικής ευστάθειας που έχει σαν σκοπό την ανάλυση της συμπεριφοράς ενός συστήματος ηλεκτρικής ενέργειας, στο οποίο συμβαίνουν κάποια γεγονότα, τα οποία επιφέρουν άρση της ευστάθειας αυτής.