Use of WOS for Trickle Loads
During trickle loading, the WOS permits Vertica to batch small data loads into larger ones for I/O purposes. Loading to WOS is significantly faster than loading to disk, but it has size constraints. To help balance speed, memory size, and data flow, trickle loading defers the work of sorting, encoding, and writing to disk. The Tuple Mover's moveout process performs those tasks in the background when a table or DML operation uses trickle loading as its load method.
WOS capacity is significantly less than ROS. When incoming data exceeds WOS capacity, Vertica spills small loads directly to disk. While no data is lost when the WOS gets full and spills to ROS, it can result in wasted I/O bandwidth. For optimal performance, follow the guidelines described in Managing the Tuple Mover to avoid WOS overflow.
If the WOS becomes full, before the tuple mover can move data to ROS, the entire load fails and is rolled back.