Replication question

We have one master site running 7.3.2.3 and 19 snapshot sites running 7.3.2.1. This implementation has been running for two years without an serious problems (believe me we have had our share of problemsthough). We are using a combination of read-only (force) and updatable snapshots. Here is the problem:
We created two read-only snapshots at all 19 sites and added them to a single refresh group. One table contains 8,000 records (parent) and the second contains 250,000 (child) on average. Every morning at 2am we delete parent records and their children that are older than three days which amounts to about 2500 in the parent table and 80,000 in the child table.
After the delete process, the refresh takes a considerable time which is expected. On four sites, the refresh job stays in dba_jobs_running for about 3 hours and then fails. The log files on the snapshot site contain the following:
ORA-12012: error on auto execute of job 93 ORA-01555: snapshot too old: rollback segment number 1 with name "R05" too small

ORA-02063: preceding line from stage_db (master site)
ORA-06512: at "SYS.DBMS_SNAPSHOT" line 271
ORA-06512: at "SYS.DBMS_IREFRESH" line 413
ORA-06512: at "SYS.DBMS_REFRESH" line 171

I am assuming that the actual error is the rollback segment at the master site.
My question is why is the rollback segment on the master site being used so extensively during the refresh of a read-only snapshot with refresh FORCE. Doesn't the snapshot site just query the MLOG$_ table for all changes and perform a complete refresh if this fails. What is being written to the rollback segments on the master site?

Please cc glenn_at_corp.home.net on any replies. Thanks,
Glenn Received on Mon Jul 13 1998 - 14:38:00 CDT