We had a very good presentation today from Kai Yu at Dell on how to implement EBS 12 on RAC.
Transcript edited to only have technical questions
$Jatin : any idea why oracle recommends to have odd number of voting disk
Dan Norris: I'm here learning, just like you. Voting disks: odd # to avoid ties
Adrian Munteanu: ASM can work with raw devices as well but actually by default it works on block devices
Sylvie Berube: why not putting concurrent managers on database nodes?
Dan Norris: Q: I've seen some EBS on RAC environments that concmgr is installed and runs on DB nodes. What's the recommended configuration and advantages/disadvantages of each?
Timothy Kriel: Good question Dan - I know what my guesses would be but I dont know for sure
Ahmad Iraky: Hi Dan, note 304748.1 has the recommendation of having Conc Processing with Application tier, historically the recommendation was to have the conc manager with DB for better performance but with the increased network speed now, it is recommended to have it with Application tier to reduce maintenance efforts
Venkata Giri Gi: Is the split configuration supported on all the platforms to support this architechture
Dan Norris: @Ahmad no such note found for me, but thx for the summary.
Sylvie Berube: this is not considered a split configuration because using 64-bit all the way. it is just standard EBS distribution. i still question the CM on the app tier. this would slow down considerably the reporting should the app tier be in a different network domain.
Dan Norris: Q - Kai: Does having RAC in 11i make for faster upgrade to R12? Logically, I would think yes, but not sure if upgrade can be parallelized to take advantage. Did you test both and compare?
Ahmad Iraky: Dan, it is internal note, needs Oracle login to view it
Dan Norris: @Ahmad Okay, hopefully the super-secret stuff can be removed so the rest of us can know the recommended config too.
Adrian Munteanu: Dan if you use 10gR2 rac for 11i then that db can be used for R12 as well without having to redo the rac migration if that was your question
mike smith: For a fresh R12 impl. can somebody confirm that Oracle Note 388577.1 was used...or was some other method used? If so, what path was used for conversion of single instance to RAC...Thanks
Dan Norris: @Adrian, not my Q. My Q is this: If I'm on 11i/single now and will upgrade to R12/RAC, should I upgrade first, then RAC or RAC first, then upgrade? What are the complexity vs. possible speed of upgrade improvements by doing RAC first?
Adrian Munteanu: Dan it doesn't matter when you do the migration to RAC
Adrian Munteanu: in the case you specified best would be if you do it with R12
Adrian Munteanu: so after you already moved to R12
Dan Norris: @Adrian, best how? Because it is less complex? Is there no performance benefit during the upgrade if we're on RAC?
Adrian Munteanu: Mike yes, note 388577.1 is the document
Adrian Munteanu: that is the only certified method
Adrian Munteanu: rac will give you no real benefit while driving the upgrade
Adrian Munteanu: the r12 upgrade consists in two big steps:
1 lay down the r12 middle tier
2. make the proper changes in the DB to cope with r12 code
Adrian Munteanu: in the last phase you run a driver via adpatch
Dan Norris: @Adrian So the upgrade is essentially single-threaded?
Adrian Munteanu: true
Dan Norris: @Adrian ok, but ick. Hope that is on the enhancement request list
Timothy Kriel: Adrian - during the aplication of the unified driver (database portion only) when upgrading to R12, I noticed that adpatch spawns many sql scripts run by the ad workers.. will they not take advantage of load balancing with RAC?
Adrian Munteanu: no as adpatch can't work with rac
Adrian Munteanu: applications is not rac aware
Adrian Munteanu: this is going to be fixed in fusion probably
Timothy Kriel: Thanks Adrian
Adrian Munteanu: nor middle tier services can fully take advantage of rac capabilites like taf
Nick Wang: @Adrian, so why RAC?
Naqi Mirza: during the upgrade one could leverage distributed ad (make use of the processors on the other rac nodes) to improve the patching time, no?
Adrian Munteanu: for example if u run rac with ebs and the rac node where the http connection is opened fails you need to refresh the page to connect to the other node(s)
Nick Wang: since neither applications or MT can take full advantage of RAC
Adrian Munteanu: there are other advantages for using rac
Adrian Munteanu: fail over for example
Timothy Kriel: Nick, during the upgrade / install of EBS, RAC is invisible, but you convert EBS to be RAC-aware, upon which EBS takes advantage of RAC
Adrian Munteanu: reduce risks
Nick Wang: @adrian, to clarify,
Adrian Munteanu: Naqi that would be an option but who will drive an upgrade in production having the users logged on the other RAC nodes?
Nick Wang: you mean AD tools is not RAC -aware , right?
Adrian Munteanu: epsecially as from 11i and R12 there are dramatic code changes
Adrian Munteanu: Nick yes that is correct
Nick Wang: ok, thanks.
Nick Wang: @Adrian, another Q
Adrian Munteanu: usually the issues occurs when you drive loadjava scripts/xmldb/xdfs which require direct database connection
and there is where the ad code fails and as a workaround you need to set the two_task to one of the rac nodes to run it fine
Nick Wang: say if you are logged on and on a form,
Nick Wang: and failover occurs,
Nick Wang: do you need to re-log in?
Adrian Munteanu: Nick i answered that, yes you need to relogin
Adrian Munteanu: the only middle tier service which can do auto failover is CM in one case only
Nick Wang: even for "query" forms?
Adrian Munteanu: when the physical name has only one virtual hostname assigned
Adrian Munteanu: yes as the query form means opening an applsyspub connection in the db to query
Naqi Mirza: adrian I just meant, during the upgrade, if spare nodes are available (due to rac), the processors available and idle on those nodes could perhaps be employed via distributed ad - there would be no users connected on any other than the processes that drive the upgrade via AD
Tao Zuo: can I use the oracle database software 10.2.0.1 that was intalled by EBS install after crs 10.2.0.1 installed?
Naqi Mirza: that said, distributed ad requires a shared apps tier
Naqi Mirza: so im not sure how that would factor in during the upgrade
Adrian Munteanu: naqi maybe i didn't understand your question here
Nick Wang: just to share my experience,
Nick Wang: in our company, we use Data guard instead of RAC
Nick Wang: which turns out to be a better option for us.
Nick Wang: We achieved TAF as well as DR solution with Data guard.
Naqi Mirza: the question i guess was spawned out of implementing rac first and then going to r12 or going to r12 and then converting to rac
Nick Wang: also license wise, save a lot of $$$
Nick Wang: sort of one stone kills 3 birds.
Adrian Munteanu: Tao out of R12 dvds you got 10.2.0.2 or 10.2.0.3 and you can you can use that OH yes for the RAC OH but there are few more steps to be done for the other nodes
Naqi Mirza: a thought is, if you implement rac first - depending on the number of nodes, you have additinal cpu processing power that you could make use of during an upgrade
Naqi Mirza: its not that the ad code is rac aware, but more a facotr of having the additional computing power available to you during an upgrade - via the nodes being available from a rac implementation
Dan Norris: @Nick RAC and Data Guard are not mutually exclusive, many sites use both. If you have no requirement for RAC and DG meets all your needs, great!
Nick Wang: @Dan,
Adrian Munteanu: I am with you that running adpatch it will mostly run the database driver but the workers will not be distributed at each rac nodes like the middle tier connection as adpatch is not rac enabled
Nick Wang: can you give out a scenario where RAC is needed that DG cannot fulfill?
Nick Wang: not to be party pooper...
Nick Wang: i found none...
Adrian Munteanu: Nick, cache fusion ?
Dan Norris: DG cannot provide subsecond failover for sessions.
Nick Wang: well, cache fusion brings problems of its own, doesn't it?
Naqi Mirza: adrian check out this note: 225165.1
Dan Norris: Scalability beyond the resources of a single node is also not possible with DG alone.
Naqi Mirza: check out the section on distributed ad - it does however require a shared apps tier
Dan Norris: @Nick I'm not saying there aren't potential issues, but DG is not a replacement or competing technology with RAC; they're complimentary.
Hien Nguyen: will Kai cover upgrade of an existing 11i/RAC to 12/RAC?
Nick Wang: Agreed.
Adrian Munteanu: i think just r12 single node to rac
Adrian Munteanu: naqi i knew the note but to be honest there is still nothing new as the tool is the same
Hien Nguyen: thanks!
Naqi Mirza: I understand, I am not saying that the code set makes use of rac per se, i am talking specifically about using thoes nodes which are available as a result of having rac for the upgrade - with respect to cpu
Nick Wang: @Dan and Adrian?
Adrian Munteanu: here is what distrib ad does:
This distributed AD feature of AutoPatch and AD Controller allows you to assign
workers to run not only on the primary node but also on the secondary nodes that
share the application tier file system.
Timothy Kriel: FYI - I ordered a new copy of R12 software, and it now comes in a 12.0.4 media pack
Adrian Munteanu: sorry Nick ?
Nick Wang: do you know any information about clustering R12 MT?
Dan Norris: @Nick, did you ask something else?
Nick Wang: yes,
Nick Wang: can you do R12 front-end clustering?
Nick Wang: to further increase availability of applications?
Adrian Munteanu: Nick was this what you were looking for? This distributed AD feature of AutoPatch and AD Controller allows you to assign
workers to run not only on the primary node but also on the secondary nodes that
share the application tier file system.
Nick Wang: say, when you apply patches...
Adrian Munteanu: ooops sorry
Nick Wang: E-biz patches...
Naqi Mirza: i think he is asking about using load balancers
Naqi Mirza: at the middle tier
Nick Wang: Let me rephrase my question:
Naqi Mirza: for clustering
Adrian Munteanu: is note 380489.1
Nick Wang: in the future release E-business..
Guest 15: In JDBC url, can we also specify failover mode like for select, # of retries and delay
Guest 15: ?
Nick Wang: Is there any attempt made to apply patches while applications is UP and accessible?
Naqi Mirza: forms in ebs are not rac aware
Adrian Munteanu: Nick that was for you
Naqi Mirza: @guest15 that is
Nick Wang: in other words,
Nick Wang: apply patches to node1 while node2 is serving requests.
Nick Wang: Rolling upgrade for functional patches...
Nick Wang: not sure if you get all that..
Adrian Munteanu: Nick and after patching node 1 what?
Adrian Munteanu: the changes will be done on node 2?
Adrian Munteanu: or bring that down for patching while node1 is serving?
Nick Wang: yes? is that possible at all?
Adrian Munteanu: it can be reached in many ways by now
Adrian Munteanu: MAA is one of the ways
Nick Wang: gtg, Arian and Dan, you guys are awesome!
Guest 15: Thans Kai, good presentation
Timothy Kriel: Thanks Kai!
Dan Norris: FYI, Steven Chan, one of the PMs at Oracle, maintains a good blog on lots of the technology issues related to Apps (11i and R12) at http://blogs.oracle.com/stevenChan/
Rahul Gandhi: It's really good infomative presentation, can v get this presentation
Jan Chen: Thank you!
Sathishkadiyala: Rahul we can get
Dan Norris: The preso slides will be posted on the RAC SIG website in the next couple of days...in the document library
Rahul Gandhi: Thank you
Sathishkadiyala: Thx Dan
Sarang Sagdeo: excellent presentation
idc-internal_u: A replay with PPT of this event will be available on OUWeb in 3 business days
Sashi Kondaredd: Thanks Kai. Excellent presentation.
Jose Valerio: Great presentation ! Actually we've customers that alreday are in RAC 10g using eBS 11i. My Q. is, for those customers do we have a study case or Metalink Note?, propably migration can differ considerably
Lei Sun: what's detailed ur on the OUWEB?
Wei Wang: Could you please explain the laod_balancing for different components of EBS like Form, report, etc. How it works?
Ignacio Ruiz: Thanks Kai, excellent presentation. Thanks Dan, Nick and all that enriched this session... was enlightening
Adrian Munteanu: Wei that's not rac but I gave a doc id earlier serving that purpose
Ali: Hi DAN. If we have 4 nodes and we have to install db tiers (in RAC env) on 2 nodes
Adrian Munteanu: note 380489.1
Erik Peterson: We have an new EBS on RAC forum on E-Business Suite on Oracle RAC
Erik Peterson: we will do followup questions here
Ali: and apps tier on remiaing 2 nodes then where we should install concurrent processing tier
Dan Norris: I think per Adrian's previous reference to an Oracle internal (not published) ML note, the current recommendation is to put concmgr in middle tier, possibly on one or more nodes dedicated to that purpose.
Adrian Munteanu: Q for Kai: do they share their RDBMS OH between RAC nodes?
Dan Norris: In your small env, maybe it should be on both of the app tier nodes...if resources are available.
Linda Su: What 's URL to download this presentation ?
Adrian Munteanu: Sorry Dan but I didn't comment on that
Wei Wang: THanks
Erik Peterson: presentation will be uploaded to www.oracleracsig.org
Dan Norris: Oh, well, someone did--thought it was you.
Erik Peterson: PDF this week
Erik Peterson: recording in about a month
Erik Peterson: any other questions for Kai?
Dan Norris: Sorry, it was Ahmad that sent this: "Ahmad Iraky: Hi Dan, note 304748.1 has the recommendation of having Conc Processing with Application tier, historically the recommendation was to have the conc manager with DB for better performance but with the increased network speed now, it is recommended to have it with Application tier to reduce maintenance efforts"
Adrian Munteanu: Kai I was talking about the RAC OHs
Adrian Munteanu: so is not shared
Kelly T: Thanks Kai. Very good presentation.
Adrian Munteanu: 10x for the answer
anil gupta: Thanks.....Kai.
John Ospino: Kai, good presentation!!! thanks
Adrian Munteanu: Another Q for Kai: did they try 11g RAC by now?
Dan Norris: Erik-- can we see if we can post the chat transcript from today's session on the RAC SIG site too? Lots of good nuggets in here today!
Dan Norris: good WP on OHs in RAC http://www.oracle.com/technology/products/database/clustering/pdf/oh_rac.pdf
Edited by: ErikP on Nov 20, 2008 11:04 AM