Sunday started off a bit oddly at registration, having done the mobile check-in, it seemed like there was quite a bit of confusion over the printing of my badge. I was given no lanyard, and no booklet of tickets for OTN night or lunch vouchers. Then shuffled over from Moscone West to Moscone South to collect materials which turned out to be a solitary appreciation night wristband.
No bag for bloggers! This lack of materials got overturned later in the day though and thankfully I was able to plead for a ticket booklet. Not quite a smooth as last year, but in the end I was grateful to have done the mobile checkin and done the badge collection early as there were massive queues later on.
Field Report: Extended RAC – Andreas Ellerhoff
A quite deserted presentation with an astongishingly formal start to the presentation from Andreas Ellerhoff he is speaking incredibly slow-ly. – though humorous in saying “hello Opeworld from Germany” – I laughed but not many others did, sounded like a reference to the eurovision song conference – I guess that might be missed by americans
Just seen a slide which looks like it’s at 6 point font. learning a lot about duetchse messe not much about extended RAC yet.
They had a distance between datacenters of around 0.7 miles. Though there were not many details regarding their network configuration.
They have a non-symmetric setup between data centers
Unbelievealbe! There really were very few interesting details in this presentation and it finished in 30 minutes
Under the Hood of Oracle Clusterware 2.0–Grid Infrastructure – Alex Gorbachev
Alex gave a great presentation to an absolutely jam packed hall.
Some very nice graphics explaining node eviction.
mentions cssd monitor to kill cssd if it’s unresponsive
if voting disk is missing both nodes will commit suicide
11gR2 brings in a member kill operation where LMON instance can request remote instance be killed via clusterware. if it fails clusterware will evict the node which ignored the member kill operation
11gR2 brings in support for ipmi to kill a node – it can power of the node.
Exclusive exadata fencing mechanism
init.ohasd run reads file: ohasdrun
ohasdstr – enable/disable
Agents get spawned
Some processes run as root, some as Oracle
olr (oracle local registry) used to locate voting/ocr disk stored asm, accessed by clusterware before asm actually starts
ocr written only by crs and only from a single master node. 11gR2 olr managed by ohasd
Alex is praising 11gR Grid Infrastructure thinks the software is much better designed than previous versions. Possible to have sophisticated policies and resource dependencies
Alex says according to the experience at pythian customers 11gR2 clustering is a more stable than 10g.
RACSIG Experts Panel
A really good discussion this one.
126.96.36.199 Quality of Service Exadata only
ASYNC Rebalance power limit has increased
realistically we see maximum of about 50-60 miles depends heavily on latency and bandwidth of two sides.
cluster health monitor installed with 188.8.131.52 crickey that patchset has more new features than you can shake a stick at – apparently this uses berkley db! You need more space for this.
Oracle local registry – olr used for oracle restart
In the evening headed to OTN night and then had a few beers with some of the crew from RittmanMead.
Was falling asleep in the above by 10 O’clock!