Replica Placement For Improving File Availability Of Medical Imaging Application On Grid

Biomedical imaging is a indispensable mean and tool for modern medical re-search and clinical diagnosis. Along with the development of medical imaging, in medical image process, the requirement of high performance computing and mass storage is highlighted. Because grid is able to provide powerful compute resources and a solution for secure storage and data sharing for for medical imaging applica-tion, it has been widely used in discipline of medical image process. However, in in real world applications, the data management of medical imaging application in grid still is not efficient enough. Many jobs of medical mimaging application in grid are failed for file transfer failures caused by file unavailability. Although optimizing replica placement can increase the file availability in grid, replication researches are seldom dedicated for file availability.Motivated by data management service on Virtual Imaging Platform, dedicate to file placement for improving file availability on grid, replication method classifi-cation, file availability model, job success probability model and metric of replica-tion strategy optimization are studied. The research result of the subject provides a theoretical foundation to data placement of medical imaging application in grid. The results of the subject are not only fit for data management problem on grid for med-ical imaging application, but also could be used on similar occasions in other grid applications. These results are also valuable to data availability in distributed system such as cloud computing.The mainly research works are shown as following:Firstly, to study the affection of data placement on file availability of medical imaging application in grid, workflows on Virtual Imaging Platform are observed, related executing data such as data placement, availability of storage element and status of jobs are extract from log files. Data analyzation shows that, appropriate data placement strategy is significant for promoting file availability on Virtual Im-aging Platform and reducing job failure rate of workflow in grid.Secondly, to show the characters of replication method in grid and its applica-tion rules, replication method taxonomy are improved and perfected. Based on rep-lication taxonomies exist, using Resource Description Framework, a replication method classification is presented. By the taxonomy, recently researches and repli-cation method required by medical imaging application on grid are analyzed. This analyzation shows that RDF kind classification of replication method can provide an observable and measurable taxonomy, which can quantify the distribution character and application rule of replication method research. Thirdly, to quantify the relationship between jobs’ success executions and file availability on production grid, file availability and job success probability is mod-eled. Based on the file availability researches before, stateless and stateful file availability model are suggested, so that job success probability is modeled. The job success probability can reflect the effecting of data placement strategies and storage element availability. Theoretical analyzation proves that, comparing with stateless model, stateful model always can show the effecting of data placement strategy and replica number. Experiments show that the job success probability model is reliable. The stateful job success probability model is more accurate than the stateless model to predict success rate of medical imaging application in production grid.Lastly, to study metrics of data placement strategy in grid，system brittleness theory is applied for analyzing replication problem in production grid. Job system brittleness entropy model is formalized as a measure function for data placement strategy. This model does not only reflect uncertainty of storage elements configu-ration, but also the job failure risk caused by file transfer problem. The model re-flects the relationship between data placement strategy and job failure risk quantita-tively. Theoretical analyzing shows that, using job system brittleness entropy as a metric of data placement strategy is feasible. Combining job system brittleness en-tropy model with job success model, the relationship between data placement strat-egy and job success probability model can be shown more clearly. This is a theoret-ical basis for further study on data placement strategy optimization in grid to resolve data transfer problem of medica imaging application in grid. The experiments show that using job system brittleness entropy as a metric of data placement strategy is feasible.