I exported a db into a replacement database recently. The original database had one datafile per tablespace, but the new one has 3 per tablespace spread over 3 different disks.
Example: One TBS had 3 x 100mB files. The Import filled one file 100%, the others: 25% and 27% (I had expected an even spread of 50% for each datafile).
Another TBS - 3 x 400Mb. Had a spread over the disks of 73% - 24% and 26%
A third TBS - 3 x 350Mb had a spread of 21% - 47% - 21%
Can anyone xplain to me how Oracle decides which disks to use? Is Import different from the normal load balancing I might expect?
Its 9.2.0.1 on Win NT4 and the TBSs are locally managed with uniform extent sizes of 1Mb
Well, no, maybe not. I'm just interested in why one file filled 100% when the other TBSs have a more even spread of data.
I might prefer to have an even spread. There's no partitioning and I might want the access to any row to be more balanced and not 'stress' any particular datafile or disk.
Oracle tries to spread the data equally among the files. Round-robin fashion.
Which is what I thought.
One issue might (or might not!) be the DB block size.
The exporting DB had a block size of 6K, the importing one had a block size of 8K. To get round this I pre-created the tablespaces in the target DB, and set IGNORE=Y so as to ignore the errors when the FULL import tried to create Tablespaces of 6k blocks.
However, the fact that the spread is not consistent and one TBS has a 100% full datafile leaves me to believe this is not the case.
Originally posted by julian
Oracle tries to spread the data equally among the files. Round-robin fashion.
From what I observed with a MOVE, it would seem that this is "simplistic" in its implementation. It looked as if, for each table, Oracle always started with the same file - so that if all your tables had only one extent, they would all be in the first file (not sure how "first" is defined).
Reply With Quote
Bookmarks