I recently attended a workshop on upgrading to Oracle 11g, held at Oracle’s TVP offices in Reading. This was a heavyweight day, from Oracle development bods, rather than a marketing led effort. Apparently they are on a global tour with their 11g upgrade workshop.
The 400+ slides are currently available for download courtesy of Oracle’s Mike Dietrich.
If you are interested in upgrading to 11g do have a look at the slides.
One refreshing blast of honesty was that they stated one of the reasons for upgrades was that people were “Forced to Upgrade due to Oracle Support policies”, that is versions go out of support quite quickly and while you can purchase extended support, at a cost over and above normal support prices this might not last all that long.
Claims that 11gR1 is effectively 10gR3, though I am sure I heard this exact claim with 10gR1 being 9iR3. Oracle are also claiming less pain migrating to 11gR1 from 9iR2 than if you go to 10gR2, saying that the optimizer is less likely to give you dodgy plans than if you’d upgrade to 10gR2.
11gR2 is slated to be released later this year. SAP will NOT certify against 11gR1 but are waiting for 11gR2. The 3 Oracle people keep saying they are keen to sign up more reference customers.
There was an interesting graph of the number of SR support calls of various versions, 9, 10, and 11 plotted over time. They all show a slow ramp up, with the 9i release, leveling off and actually declining, while the 10g in the leveling off phase now.
Emphasising the need to have a timezone of version 4 before attempting an 11g upgrade – note the upgrade scripts will error out if you do not upgrade the timezone, utlu111i.sql which you have to run in advance of upgrading, shows you if you need to upgrade your timezone.
Length of time the upgrade takes is independent of the quantity of data in your database, but does depend on the number of objects you have, the number of synonyms you have created and the number of different features you are using.
Top tip is to ensure you have as much information on your pre-upgraded database performance as you can possibly get, e.g. extract your awr data.
Apparently one customer saved 1.5 hours out of 6 hours, so a 25% saving by making sure they had up-to-date dictionary stats. So create your dictionary stats before you attempt an upgrade.
Apparently a large auction site is running with commit_write set to nowait to ensure there is less delay with transactions – of course this can lead to portential data loss.
Some companies are using Transportable Tablespaces for very fast upgrades, including patchsets. The idea is to make datafiles visible to two servers then the downtime required is only making the tablespaces read-only and exporting the metadata, seen customers have downtime measured in minutes.
Another idea for fast upgrades was using logical standby including transient logical standbys.
Mentioned a view I had not heard off before, V$SYSAUX_OCCUPANTS . This shows you what features are using the SYSAUX tablesspace and how if at all you can move this to point to another tablespace.
APEX no longer needs separate http server, but can connect via embedded gateway
This was an excellent jam packed day and FREE as well – you don’t get to say that all that often in relation to Oracle! Clearly Oracle are keen on getting customers upgrading to 11g, and I did wonder if take up of 11g is perhaps not as strong as Oracle would like.