From April 10th to April 14th 2016 many Oracle professionals congregated in Las Vegas. The three big U.S. user groups, the IOUG, the OAUG and the Quest User Group once again formed one of the best Oracle Conferences in the world: COLLABORATE16. I was honored to be there as a speaker, and glad to perform two lectures in front of packed rooms (downloads here). Thank you all who made it possible and who attended!
But this post is about what I took home from COLLABORATE16. I never take the time to write down everything. But wherever I go, I take notes what to look at when I am back. And as often, I’d like to share my very personal best-of list of souvenirs from Las Vegas. It’s ordered by time – first session first, and all of them are top notch state-of-the-tech speakers, no quality comments necessary. It’s always a privilege to meet you, thank you all for sharing your experience!
Linux/Unix tools for the DBA
by Tim Gorman
- sar -w context switching
- sar -W (virt mem swapping)
- sar -B (virt mem paging)
- sar %steal = time stolen by hypervisor
- sar = sadc (collector) + sadf (formatter)
- mpstat (per processor stats)
- mpstat column icsw (involuntary context switches)
- ipcs interprocess comm data structures or resources
- shared mem
- semaphores
- message queues
- Oracle sysresv = IPC resources for $ORACLE_SID
- pmap = process mem mapping display
- pmap -x for all oracle processes / max of shared plus sum of the privat comp = mem oracle uses
- jstat = Java VM mem usage displays
- jstat Stop The World STW means cleaning up old objects due to dependencies / inheritance chains (vs. young ones w/o stopping)
- jstat -gc (FGC column -> STW events since process start)
- dtrace – very powerful performance analysis tool
- adb – obtain stack trace from core dump
- Oracle alter system dump
- Oracle sosreport
- Oracle OSwatcher
Oracle on Windows
by Mark Sorger
- Book Windows for the Oracle DBA
- WIN derives from VAX/VMS
- Stopping Oracle Service stops DB, but stopping DB leaves Service running
- sc service control utility – powerful
- Logging Oracle goes to Application event log facility – Also OS audit!
- %LOCAL% on Windows works like $TWO_TASK on Unix
- Shutdown server does not mean shut down DB cleanly ORA_SID_SHUTDOWN does not work as Documented.
- Workarounds available
- bcdedit utility (e.g. /3GB switch)
- blat.net email spooler for WIN
- taskkill
Whats next with RAC (12.x)
- 12.2 Instances on Leaf Nodes
- Oracle RAC 12.2 will have Service Oriented Buffer Cache Access
- Rapid Home Provisioning for Grid Homes to come with 12.2 – deploy fully patched homes
- Oracle 12.2 ASM Service can centralize management of shared disks for all RAC systems in the domain
- Member Cluster <-> 12.2 Storage Service IO Service talks in SQL net like FlexASM plus data payload
- version 12.2 will allow PSU installs with Universal Installer
- ASM Flex Diskgroup Filegroup allow to group by database, so split mirror per DB on ASM level possible
- Autonomous Computing / Auton. Health Framework
- Maximum Consolidation Architecture is MAA + gudance for consolidation
Incident analysis using ASH data
- ASH ring buffer for 1h in Shared Pool -> 10% to DBA_HIST…
- make ASH report per SQL_ID part of the default procedure when optimizing
- BloodHound by OraPub
- bhtick (possible in SQL Developer?)
- bhtl (Timeline)
- bhvisual (Data dump for R or Mathematica), see BloodHound README for step-by-step
- Check out “R” Toolkit (available for Mac)
RAC Best Practices
by Anil Nair
- Batch Upgrade in parallel for “middle” nodes
- Flex Cluster switchover load bump?
- Read Only/Mostly workload on Leaf Nodes
- Oracle 12.2 Flex Diskgroup
- ASM VAM vs PAM (12.2 prefers VAM)
- Service-oriented Buffer Cache Access incl. pre-warming cache by service name
- Node eviction: So far Lowest node number (changed!)
- Node Weighting: Workload (majority of services or overrule w/ critical service)
- NUMA aware to avoid QPI hops
- RDMA will be available
- Hang manager
RAC and InMemory
by Markus Michalewicz and Andy Rivenes
- Inline LOB works w/ IM, outline not – obviously
- InMemory Distribution by Partitions / Subpartitions
- Keeping SMUs up to date is like keeping two Tree indexes up (Oracle Benchmark)
- In 2-node RAC you use SGA size 2*((4/2)*1.1) for additional block management
- Auto-DOP on RAC ensures to have one PX slave on each instance
- IMCU Redundancy only in Exa
- Service prefered per instances (srvctl) so nice