Question about replication and Mat.View Logs

[efrijters]

Hello,

I have a db with snapshot replication. It refreshes data from a production db with in total of 9 Gig of data. The materialized views have indexes, which cause the replication to take a lot of time (almost a day).

When I noticed the long refresh operation, I created materialized view logs on the production db in a separate tablespace. I altered the refresh method on all materialized views to FORCE (it will try a FAST refresh first and if that fails it performs a COMPLETE refresh - I need a FAST refresh, because that uses the MV Logs.)

The MVLog-tablespace is now 145Mb and I started a new refresh action on the replication db. My idea was that reading the 145Mb of data must go fast, but now I am 90 minutes underway but it is still refreshing.

I see that the tablespace that holds the MVLog-data, slowly decreases in size. (now 100Mb, cpu is 95% idle)

Why is it so slow? Is it because of the indexes (because there are a lot)? Is it better to remove all indexes, refresh the data and later on rebuild the new indexes?

You're help is appreciated!

Erik

[efrijters]

Supplement:

This problem is mentioned in Oracle bug 2044102.
This bug is solved in version 8.1.7.3

We are running 8.1.7.1.4

Case closed I guess...

Maybe this info can help some of you

Erik