Re: Multiple imports

On Oct 23, 12:29 pm, "astalavista" <nob..._at_nowhere.com> wrote:
> Hi,
>
> Is there any danger to run multiple imports
> to accelerate an import (with default option: ignore=n )
>
> Thanks for your lights

No problem, this is "poor-man's parallelization." Except you might have to use ignore=y, since the first thing is likely to be object creation.

You need to test, obviously there could be more contention under some circumstances. And it depends on exactly how you do it, if you simply start 3 identical imports, two of them might bog everything with the I/ O from constraint violation errors into the error log. On the other hand, if you are using 3 different exports with the QUERY qualifier to avoid overlap, through pipes into imp, your bottleneck may very well be different than a straight imp, and perhaps something you can help (for example, if you are going over a network it may make sense to both compress each pipe and temporarily put in a faster private direct network connection). There also may be freelist contention, or other oddities having to do with how blocks are loaded that may show up. Or you just may run into db writer or disk write limitations. The basic concept is easy enough to test that it is worthwhile to try it. Any problems may indeed be educational for us all.

Also, some circumstances may favor DIRECT usage. Others may favor CTAS over a dblink. Others you may want to wind up with the data contiguous - I believe Tom Kyte may disagree with that, by the way. Again, only testing your particular situation can you be sure if there is an effect worth worrying about.

It's nearly always quicker to build the indices and statistics afterwards, though I've found it is often good enough to do that as part of the imp, depends what your requirements are - if overnight is better than babysitting for 3 hours, for example.

jg

--
@home.com is bogus.
http://www.signonsandiego.com/uniontrib/20071023/news_1b23tourism.html