Fig 1.
A simplified view of openEHR multilevel model.
Archetypes are created from the building blocks of the RM to express domain concepts which represent valid data structure in the RM. By combining archetypes we can generate templates that may be used to generate forms or messages.
Table 1.
Size of Files and Databases (GB).
Fig 2.
Boxplot of the response times for all queries grouped by databases and datasets.
The dots represent outliers, which result usually from: i) faster response for some queries due to indexing in Berkeley and eXistdb, ii) and slower response for the first time each query is submitted to the databases. The y-axis is in the logarithmic scale. MySQL is the original relational database. BaseX, eXistdb and Berkeley DB are the XML databases and Couchbase is the JSON database.
Table 2.
Descriptive Statistics for the Boxplots of Fig 2.
Fig 3.
Boxplot of the response times for the sus100k dataset and one of the queries.
The boxplots are grouped by database and date interval. The y-axis is in the logarithmic scale. The outliers above the median are due to the fact that usually the response time of the first submissions of the query to the database is higher than the others. The tests were performed on a single computer. MySQL is the original relational database. BaseX, eXistdb and Berkeley DB are the XML databases and Couchbase is the JSON database.
Fig 4.
Boxplot of the indexing time for each dataset in Couchbase.
Each boxplot contains 8 indexing measurements for each query. The outliers are due to the fact that usually the indexing time of the first query submitted to the database is higher than the others. The indexing times for the sus10k dataset were smaller than 1 s and were not shown. The measurements were performed on a single computer.
Fig 5.
Median indexing time plotted versus dataset size for each cluster configuration.
The y-axis uses a logarithmic scale. The indexing time was not measured in the 1-node cluster for the sus4200k dataset. In the 4-node, 8-node and 12-node clusters, the median indexing times for the sus10k dataset were rounded to 1 s and overlap in the graph.
Fig 6.
Scatterplot of the median response time against the result set size for each cluster.
A regression line is plotted for each cluster.