The principle is exactly the same as with CTAS. Each slave process allocates a temp segment and insert data into it. At the end, the coordinator process mereges those allocated temp segments into table's extents, trimming the last temp extent of each slave process.
For example, let's say your degree of parallelism is 4. Each new extent of the table ought to be 5M in size. Suppose each of those 4 slave processes allocated 5 temporary extents during insert. 4 of those eextents are totaly packed, while the last one is probably not. So at the end, the result will probably be 16 new extents of 5MB each and 4 new extents with sizes ranging form anything between couple of KB and 5MB.
This is how parallel loading is behaving since 8i (or maybe since 8.0, can't remember).
Last edited by jmodic; 06-18-2004 at 07:39 AM.
Jurij Modic
ASCII a stupid question, get a stupid ANSI
24 hours in a day .... 24 beer in a case .... coincidence?