Chris, You can do some optimization on streams side like parallelizing apply process and tuning network parameters and upping the pool size but i do not know what will be the latency for 40-50 GB per hour of redo especially on a 1 Gbit link. Did you guys consider storage replication technologies over WAN? I really do not know their latency, but i want to see others comment on this. You have some interesting conundrum here given the link speed. Thanks Deen On Wed, Apr 28, 2010 at 4:26 AM, Chris Dunscombe <cdunscombe@xxxxxxxxx>wrote: > Hi, > > We're just looking at migrating a large database (> 20TB) with as little > downtime as possible. Here's the main information: > > - Oracle RAC 10.2.0.4 64 bit > - Source Windows 2003 (Little Endian) at site A on HP XP storage > - Target AIX (Big Endian) at site B on HP XP storage > - WAN network link is 1 Gbit > - Distance between sites is 10 - 20 miles > - Database is a Data Warehouse, 22 TB with around 40-50 GB redo per hour > during peak load periods > > So any ideas would be much appreciated, especially with streams and would > it have any chance of coping with a big database and a heavy insert load. > > Thanks, > > Chris > > > ------------------------------ > *From:* Ilmar Kerm <ilmar.kerm@xxxxxxxxx> > *To:* Oracle-L Freelists <oracle-l@xxxxxxxxxxxxx> > *Sent:* Thu, 22 April, 2010 17:15:12 > *Subject:* Re: Cross-platform (different endianness) Migration Experiences > > On Thu, Apr 22, 2010 at 5:40 PM, fmhabash <fmhabash@xxxxxxxxx> wrote: > > Just wanted to tap into this group resources for real-life experiences > > migrating large databases across OS platform and across different > > endianness. I'm looking for ... > > Three 1tb databases on 10.2.0.4 from solaris sparc to solaris x86-64. > We used expdp/impdp with streams. > > For data pump we used 15 parallel threads, export took about 2h and > import took about 15h to finish and after enabling streams it took > about 2 days to catch up (quite heavy OLTP database). The actual > switch from one database to another was completed in about 1 hour > maintenance window, that included also configuring Streams the other > way around for rollback scenario. > > Streams is quite pain to set up, but once its running it works. So > reserve quite a lot of time for getting to know Streams and all the > necessary migration steps. > And need to monitor Streams latency closely after it has catched up > with the main database. For example we discovered a big nightly > transaction that deleted 12 million rows and that completely locked up > Streams apply for a few hours. > > > -- > Ilmar Kerm > -- > //www.freelists.org/webpage/oracle-l > > > >