Because all metadata is held in memory,
and because all updates to the metadata are single-threaded, you do
not have to configure input/output contentions. Queries cause
I/O activity only when the metadata
is initially loaded into memory based on type. For example, when a
user requests metadata for a specific column, metadata for all columns
is loaded into memory. No additional
I/O for column metadata queries
will occur until the server is paused and resumed or stopped and restarted.