Well I had a bit of a leisurely morning, thought I would skip the thoughts of Ian Smith – it was bound to all be about “fusion” anyway. These are the notes I made during the presentations I attended on the first day:
Phil Grice RAC V Dataguard
Interesting demo of the dataguard broker, and it did seem to do what it said on the tin. There was a demo of fast-start failover and it seemed pretty quick at performing the failover. Though note that in 10g you must be using maximum availability for this, though there is the prospect of doing fast-start failover with maximum performance in 11g.
Perhaps the best thing about the demo was the reinstating the failed primary as a standby using flasback database, the broker did this automatically once the failed primary was brought back up – very impressive and one of the few arguments for using flashback on a primary – hmm. what performance impact and how reliable is it really?
Jonathan Lewis Playing Russian Roulette
Several examples from the forums where people were asking for one line solutions without presenting sufficient data.
One example was using optimizer_cost_ind_adj which obviously affects all queries, jonathan showed a query that by default used a good index but when the paramter was set used the bad one.
how to stop chasing silver bullets:
• put the data in the right place
• build appropriate indexes
• avoid labour intensive operations
Andy Bryant Building and running a 60TB standby
created standby using tapes – LTO2’s took 3.5 days to backup and 6 days to recover
veritas 5.1 failing on around 50 datafiles (out of 1500) – fixed in 6.0
archive log gaps restore from tape?
workaround for gaps is to kill ARCn on primary!
Backup & Recovery roundtable
Surprising that some people are using cold backups even though they are running in archivelog mode. Media management and rman retention policy could be in opposition. Interesting that around 50% were using a recovery catalog, this seems to have it’s own issues. 11g parallelize backup of individual files (cuts up a datafile into pieces so more than 1 channel can backup the same datafile).
Thomas Pustynen rman incremental backups
V$datafile_backup keeps track of upto what scn a backup relates to. block change tracking keeps a bitmap of changed blocks. You have to specify a datafile to be used to tracking the changes. Incrementally update image copies, can switch to image copy instead of restoring – requires enough disk space. It’s very easy to switch to the copy and recover this copy before bringing the copy online.
Larry Carpenter Creating a physical standby – best practices
Interestingly quite a large percentage of people were using the broker/EM to manage their standby, but only a tiny fraction used EM to actually create their standby. Larry argued that it was more sensible to use EM to create as it’s easier, but I profoundly disagree with this, as it might be easier to create but I bet it gives you far less understanding of what make a standby tick.
Overall in truth no great revelations, but the roundtables are a great idea and I hope to be at the HA one on the Tuesday. It was kinda nice being able to get into the speaker lounge (well useful being able to get some power into the MacBook Pro). As of circa 9:30pm no sign of baby #2 so it does look like I’ll be doing my dataguard after all!