Re: Slow Import

In a valiant and sublime effort,koert54 frowned, dipped a thumb in soot and doodled:

> Nuno - 30GB 4.5h - not bad - maybe this is something for you
>http://sourceforge.net/projects/pepi
>

Thanx. I've seen it. I don't need any faster at the moment, but am prepared for when it's needed. Got a script that uses a file with all the table names (out of SQL*Plus) and generates multiple imp streams for the data which I can then run as much in parallel as I want. Can't go for broke because in this case I've only got 4 CPUs to play with and 4 imp sessions are enough for saturation.

With that, the import time goes down to just under 3 hours. Including the indexes. At the moment, I'm not using it. But as the db keeps growing, it will come in.

One thing I've noticed is that some older unix versions cause all sorts of locking problems when two or more imp processes read off the same dmp file. I've got a funny feeling older versions of imp opened the dmp file in some funny mode that prevented parallel executions. Haven't had this problem yet with 8.0 and 8.1 but in 7 and earlier it was a bitch.

Another thing that I forgot to mention: in general with this separation of import runs, a large rollback segment is not needed at all. Inserts use bugger all rlbseg if indexes, triggers and constraints are disabled.

Cheers
Nuno Souto
nsouto_at_optushome.com.au.nospam Received on Sat Nov 03 2001 - 04:49:07 CST