2 Replies Latest reply on Nov 1, 2012 1:59 PM by Richard Harrison .

    Question on NETWORK_LINK import


      We are going to move 4 schemas with a combined size of 130gb to another DB using the NETWORK_LINK impdp. We are planning to use
      PARALLEL = 5
      Has anyone had any detrimental effect in using PARALLEL for network import using NETWORK_LINK ?

      Documentation says that tables with LONG columns cannot be imported when using NETWORK_LINK. Have you guys faced any issues with any other data types ?
        • 1. Re: Question on NETWORK_LINK import
          No one can possibly answer the question you've asked nor would any answer have even the slightest value.

          Your ability to choose parallel paths depends on (A) CPU threads available and (B) CPU threads utilized.

          You have 4 threads you are going to have a hard time allocating 5
          If you have 64 threads and 63 are already constantly in use you are going have an experience with slow.

          That someone else had an issue is, as I indicated above, irrelevant just as the fact that I have never had one doesn't mean you won't.

          Read up on parallel processes, what they are, how they work, and then look at your hardware and operating system.
          • 2. Re: Question on NETWORK_LINK import
            Richard Harrison .
            I have really good experiences migrating 10g to 11g over database link. As you say long and long raw do not work but everything else seems fine ( lobs etc). Make sure you are patched up as high as you can go as there are lots of datapump bugs generally - not just with db links.

            It's much more efficient that having to go out to a file.

            Parallelism will do 5 separate tables at the same time - it does the biggest 5 first. Indexes will then be built with parallel 5 after the tables.

            Just make sure streams pool and PGA are plenty big enough both sides.