A rainy day london for the RAC & HA Sig.
First up as per usual was the survey of who is using what, Julian dyke
9.2 a few
10.1 a few
10.2 almost eveyone
11.1 two people in production
solaris on sparc around 10 a similiar number to AIX and 32 bit linux
The majority on linux x86-64. A little sprinkling of windows. No one on windows itanium. One person shouting out for VMS.
majority on 2 nodes. a few with 3, and a few with 4. A few with 3 One guy with 10 nodes and one with 12.
everyone on SAN hardly any with NAS.
Going on for 1/2 or more using ASM, a sprinkling on OCFS.
A good lot running physical standby with RAC and a surprising sprinkling using logical standby.
Only around 20% applying critical patch updates
Phil Davies Support Update
Interesting problem about IOT corruption when doing a 9i to 10g upgrade Metalink note: 471479.1.
Apparently Oracle developers now have to justify why a patch cannot be rolling.
Joel Goodman Managing Sequences in RAC
In some ways the name sequence is a misnomer as they are not actually guaranteed to be sequential, there may be gaps, they may even wrap around. They need not start with 1 or increment by 1.
Sequence cant be rolled back.
SQ Enqueue is used during cache replenishment.
gaps occur due to sql being rolled back (seuence number can’t be rolled back it’s an autonomous transaction to get the sequence number)
metadata ages out of row cache
In RAC there is the option of order or noorder. This controls the ordering of generated numbers. In increasing levels of performance:
nocache and order
nocache and noorder
cache and order
cache and noorder
With cache in RAC each instance caches it’s own set of numbers
Simon Haslam Fast Connection Failover
FCF basically means subscribers (app servers) get notified of events occuring on the db tier.
Possibility of using JDBC driver without an Oracle client with FCF.
Simon had a demo of a rac node failing and the app server still being used for queries without any intervention. JDBC driver has some methods that can be used with FCF.
Be interesting to see if this will work with tomcat, but as it’s at the jdbc level there is no reason it should not.
Mo Beik Sun & HA
This was a marketing talk, seems somewhat outdated, I’m sure there are lots of customers using Sun, Oracle Veritas, but really with 10g (&11) oracle are out to kill veritas anyway.
Claimed Oracle 11.2 was slated for end of year 2008. Hmm, is this going to be announced at Openworld?
He showed some graphs comparing ASM with SharedQFS, and as the number of nodes in a cluster increased (6, 7 & 8 nodes) SharedQFS seemed (in these sun tests) to actually outperform ASM. At lower number of nodes there was not really anything in it.
Harald Van Breederode Client Connectivity in a Dataguard Environment
how can one prevent clients from connecting to the wrong database in a dataguard environment?
Harald’s idea is using database services and after startup triggers to set services based on events happening on the database, e.g. if the database becomes read only, then a service is started that reflects this. It is now just a case of ensuring that clients connect to the appropriate service. This can be used in case of switchovers or failovers.
This was a really clever way of ensuring application connectivity in a dataguard environment.
Martin Bach Oracle RAC on Oracle VM
paravirtualisation vm’s do not (necessarily) simulate hardware
vm performance closer to physical hardware but requires kernel changes.
This seemed like a great way of having a play with RAC without having to have additional hardware.
Julian Dyke’s Tour
Julian has been presenting in the far east and Australasia. He has encountered quite a few CRS corruptions he emphasized how important it is to backup your OCR automatic backups.
As usual highly worthwhile event.