12 Replies Latest reply: Jun 20, 2014 12:09 PM by 877440 RSS

    any major differences of exadata database, listener, processes than regular RAC environment?

    877440

      I would like to ask for any input regarding any major differences of exadata database, listener, processes than regular RAC environment.

       

      I know now the exadata not only have SCAN listener, but a lot other listeners. can expert here provide more details?

       

      Thanks

        • 1. Re: any major differences of exadata database, listener, processes than regular RAC environment?
          Marc Fielding

          Hello user301176,

           

          The only major difference in listener/process architecture on an Exadata-hosted DB that I can think of is the addition of a diskmon process in the grid infrastructure.  The official documentation (DB, GI) does a pretty good job of enumerating and describing the processes.

           

          As for listeners, the SCAN and local listeners are the same as other 11gR2 RAC systems configured with SCAN.  The SCAN whitepaper has a great overview of how SCAN works, as well as some of the gochas that come when involving older Oracle clients.

           

          HTH!

           

          Marc

          • 2. Re: any major differences of exadata database, listener, processes than regular RAC environment?
            877440

            Thanks.  For listeners, there is listener for Infinite band?

             

            Another question is how you guys deal with the default sample database oracle engineer created? delete it?

            • 3. Re: any major differences of exadata database, listener, processes than regular RAC environment?
              robinsc

              Hi ,

              by default there will not be any listener on infiniband . If you need one ( i.e you have exalogic, exalytics or conventional servers with infiniband cards ) that will connect overr the infiniband private subnet you need to configure an additional listener .

              see Configuring Oracle Exadata Database Machine for Use with Oracle Big Data Appliance

              or http://vijaydumpa.blogspot.in/2012/05/configure-infiniband-listener-on.html  forsome examples how todo so.

              Howver Normally you do not need the configure additional listeners.

               

              You have 3 scan listeners per cluster and one node listener per compute node.

               

              We kept the default database but created our own ones. If you have platinum support they generally will patch the sample database home. So if you don't want them patching you databses you can keep the sample database home for them to patch and install your own databases in another home. This is especially useful if you are hosting multiple e-business instances.

              • 4. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                Kasey.Parker

                user301176 -

                 

                The base onecommand install on Exadata does some base setup for the listener on Infiniband - called an SDP listener. This is probably what you are seeing. This can be used for communication between databases on Exadata or, as mentioned by robinsc,  for communication from other systems, e.g. Oracle Engineered systems, that are connected via infiniband and can use the SDP protocol. You can follow MOS note: "Setup Listener on Infiniband Network using both SDP and TCP Protocol (Doc ID 1580584.1)" as a guide to help setup your SDP listener and connections if it applies to your situation. This document will also help you verify what has already been setup on your Exadata as part of the base install.  Setting this up allows database connections to use the SDP protocol over Infiniband instead of TCP and depending on the network traffic between client and database can provide significant performance benefit.

                 

                Regarding the initial database setup as part of the onecommand base install on your Exadata. Robinsc had a good use case... another common use case is to use that database to manage your DBFS volumes, if you choose to use DBFS. Really whether you keep and use that database or drop it is completely up to you.

                 

                Thanks,

                Kasey

                • 5. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                  877440

                  Thank you all very much for your input.  Those are really helpful. 

                   

                  I have some more questions here.  We will use zfs to do backups, is that possible for the initiband listener to communicate with ZFS?  where I can find documents for zfs?

                   

                  Plus how do I know we have exalogic?  I dont think we have one, but is there a way to check to be sure?

                   

                  So basically after oracle engineer installed onecommand and created sample database, as oracle dba, we can dbca to create a database just like regular RAC environment?  There is nothing specific on exadata perspective?

                   

                  Thanks again in advance.

                  • 6. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                    877440

                    Another question is do I have to configure exadata specific parameters in order for all those features of smart scan, storage index, comrpession, etc work?

                    • 7. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                      Kasey.Parker

                      We will use zfs to do backups, is that possible for the initiband listener to communicate with ZFS?  where I can find documents for zfs?

                      The SDP (Infiniband) Listener is for client connections connecting to the database on Exadata over the SDP protocol on the Infiniband network. You should use the Infiniband network to connect the ZFS to Exadata and mount the ZFS shares to the Exadata compute nodes using DNFS; but this doesn't use the SDP listener because there is no database connection originating from the ZFS you are just backing up your databases to it. See the "Oracle ZFS Storage: FAQ: Exadata RMAN Backup with The Oracle ZFS Storage Appliance (Doc ID 1354980.1)" MOS note for good information and references regarding backing up from Exadata to ZFS.

                       

                      Plus how do I know we have exalogic?  I dont think we have one, but is there a way to check to be sure?

                      Exalogic is another of Oracle's engineered systems. It contains integrated compute, network and storage - similar to Exadata - but is used for running application environments, specifically for fusion middleware (e.g. Weblogic) applications, instead of databases. It uses a ZFS storage appliance for the storage, as opposed to the storage cells on Exadata, and for virtualized environments uses the Exalogic Elastic Cloud software. You would know if you have one.


                      So basically after oracle engineer installed onecommand and created sample database, as oracle dba, we can dbca to create a database just like regular RAC environment?  There is nothing specific on exadata perspective?

                      Yes... Oracle on Exadata is still the same RDBMS - same Oracle Enterprise edition with the RAC option. The Exadata difference comes with the hardware integration and the storage cell software... the database software is the same.


                      Another question is do I have to configure exadata specific parameters in order for all those features of smart scan, storage index, comrpession, etc work?

                      Some Exadata features are mostly "black box" and work without any configuration others may take some configuration or tuning to take advantage of. For example, storage indexes are created dynamically on the storage cells... there is not much you can do to control them. For smart scans you need full table/index scans with direct path reads. So you need to tune for direct path reads. HCC requires setting up your tables/partitions to compress at one of the HCC compression levels and using direct path loads.

                      • 8. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                        877440

                        Thanks a lot. Those information really helpful.

                         

                        Regarding indexes, do we have to drop all the indexes in order direct path reads kick in?

                         

                        What are the real case scenarios you guys used? What is the best practices?

                         

                        What about parallel practice?  and IORM/DBRM?

                        • 9. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                          Kasey.Parker

                          All good questions... welcome to the world of Exadata!

                           

                          These are questions that could get into a lot more detail and discussion than a forum post. At a high level, you definitely don't want to drop all indexes on Exadata. However, you're need for indexes and indexing strategy will change on Exadata. After moving a database from a non Exadata to Exadata environment you are most likely over indexed. Indexes used for true OLTP transactions - seeking one or very few rows out of many will usually still be faster with an index. Indexes used to avoid a percentage of the records but still returning many records can often be dropped. The question on indexes depends on the nature of the workload and your application. If you have control over the indexes then test your queries and DML with the index(s) invisible. Check your execution plan, io_cell_offload columns in v$sql, smart scan wait events to make sure you are getting smart scans... and see if the smart scan is faster than using the index(s). The real time SQL monitor is a great tool to help with this - use dbms_sqltune or grid/cloud control.

                           

                          Parallelism is a great tool to help further speed up queries, and direct path load operations, and can help prompt smart scans... but use of parallelism really depends on your workload and needs to be controlled using DBRM and the parallel init parms, possibly even using parallel statement queuing, so it doesn't overwhelm your system and cause concurrency problems.

                           

                          If you have a mixed workload environment or are consolidating databases on Exadata then my opinion is IORM plans definitely need to be put in place.

                          • 10. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                            877440

                            Thanks.  I really appreciate all of you taking time answering my questions.  My last question here is how do you guys test all above performances on Exadata?  Do you gather majority queries to test on exadata machine after database is built and data populated?  or is there a tool or any cases we can use ? 

                            • 11. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                              Kasey.Parker

                              You can do some higher level proof of concept/value type testing by moving selected tables over and testing a subset of queries. However, before migrating your production environment you should conduct load and regression testing of a cloned production environment on Exadata - just as you would with any re-platforming. You'll want to gather performance benchmarks from your current environment and compare with Exadata. RAT is a great tool to help and be thorough with this if you own the license for it. There are other 3rd party testing tools on the market as well. You'll also find great benefit in the Exadata plugins for cloud control to help with monitoring from database down through the storage cells; and you can use AWR compares and the real time SQL Monitor mentioned in my previous post to help with testing and comparing as well.

                              • 12. Re: any major differences of exadata database, listener, processes than regular RAC environment?
                                877440

                                Thank you all. It is very great for your help.