Space allocation on 11g R2 on multiple data files in one tablespace

Question

hello
if the following is explained in Oracle 11g R2 documentation please send a pointer. I cant find it myself right now.
my question is about space allocation (during inserts and during table data load) in one table space containing multiple data files.
suppose i have Oracle 11g R2 database and I am using OMF and Oracle ASM on Oracle Linux 64-bit.
I have one ASM disk group called ASMDATA with 50 ASM disks in it.
I have one tablespace called APPL_DATA with 50 data files on it, each file = 20 GB (equal size), to contain one 1 TB table calll MY_FACT_TABLE.
During Import Data Pump or during application doing SQL Inserts how will Oracle allocate space for the table?
Will it fill up one data file completely and then start allocating from second file and so on, sequentially moving from file to file?
And when all files are full, which file will it try to autoextend (if they all allow autoextend) ?
Or will Oracle use some sort of proportional fill like MS SQL Server does i.e. allocate one extent from data file 1, next extent from data file 2,.... and then wrap around again? In other words it will keep all files equally allocated as much as possible so at any point in time they will have approximately the same amount of data in them (assuming same initial size ?
Or some other way?
thanks.

Answer 1

On 10.2.0.4, regular data files, autoallocate, 8K blocks, I've noticed some unexpected things. I have an old, probably obsolete habit of making my datafiles 2G fixed, except for the last, which I make 200M autoextend max 2G. So what I see happening in normal operations is, the other files fill up in a round-robin fashion, then the last file starts to grow. So it is obvious to me at that time to extend the file to 2G, make it noautoexented, and add another file. My schemata tend to be in the 50G range, with 1 or 2 thousand tables. When I impdp, I notice it sorts them by size, importing the largest first. I never paid too much attention to the smaller tables, since LMT algorithms seem good enough to simply not worry about it.
I just looked (with dbconsole tablespace map) at a much smaller schema I imported not long ago, where the biggest table was 20M in 36 extents, second was 8M in 23 extents, and so on, total around 200M. I had made 2 data files, the first 2G and the second 200M autoextend. Looking at the impdp log, I see it isn't real strong about sorting by size, especially under 5M. So where did the 20M table it imported first end up? At the end of the auotextend file, with lots of free space below a few tables there. The 2G file seems to have a couple thousand blocks used, then 8K blocks free, 5K blocks used, 56K blocks free, 19K blocks used, 148K free (with a few tables scattered in the middle of there), 4K blocks used, the rest free. Looking at an 8G similar schema, looks like the largest files got spread across the middle of the files, then the second largest next to it, and so forth, which is more what I expected.
I'm still not going to worry about it. Data distribution within the tables is something that might be important, where blocks on the disk are, not so much. I think that's why the docs are kind of ambiguous about the algorithm, it can change, and isn't all that important, unless you run into bugs.