Disk space fragmentation is the proliferation of small and unusable gaps. This problem is considered within the context of IBM's VM operating system. Facing this problem, researchers resorted to using algorithms based on memory management techniques, such as placement strategies (first-fit, best-fit, and worst-fit). Solutions based on these algorithms do not yield optimum results. Furthermore, in many cases, their efficiency and cost effectiveness are questionable. This work proposes a new method to address the disk fragmentation problem using improved I/0 techniques. Optimum results and efficiency are some of the qualities that contribute to the superiority of this approach. The algorithm was implemented and test figures were compared with calculated figures. The results clearly favor this new algorithm. / Graduation date: 1991

Made available in DSpace on 2016-08-29T15:33:23Z (GMT). No. of bitstreams: 1 tese_9278_Ata 337º.pdf: 642020 bytes, checksum: 75b1ea9fbbced5ea1bee7a6056858c98 (MD5) Previous issue date: 2015-10-13 / Este trabalho busca melhorar o processo migratório de máquinas virtuais implementando um escalonador de migrações baseado em dados provenientes de uma caracterização da carga de trabalho em uma nuvem privada. As migrações são necessárias para balanceamento de carga entre os hosts, contudo podem impactar negativamente no desempenho dos serviços. Por meio da caracterização empírica da nuvem estudada, foi observado uma grande desproporção entre o tempo de uma migração simples (menor tempo) e uma simultânea (maior tempo). Com base nesses dados, foi proposto um algoritmo para realizar o escalonamento das migrações, de forma serial (removendo a simultaneidade) e fundamentado no Shortest Job First. Os resultados do escalonador mostraram-se melhores em quase sua totalidade quando comparados ao arcabouço tecnológico base deste trabalho e o VMware. Foi possível reduzir o tempo médio e a quantidade de migrações ocorridas. / This work seeks to improve the virtual machines migration process implementing a migration scheduler based on data from a workload characterization in a private cloud. The migrations are necessary to load balancing among hosts, nevertheless they can negatively impact the performance of services. By empirical characterization of the studied cloud, it was observed a large disproportion between the time of a simple migration (shorter time) and a simultaneous one (longer time). Based on these data, it was proposed an algorithm to do the migrations scheduling, in a serial way (removing the simultaneity) and grounded on Shortest Job First. The results of the scheduler showed up better almost in its entirety compared to the VMware, it was possible to decrease 33% in the average time in 93% of the occurred migrations (total of 765 migrations).