8 Replies Latest reply on Jun 3, 2013 7:25 AM by 1010036

    Does all voting disks in n node rac have same data

    1010036
      Hi,

      I heard that Oracle/css access more than half of the voting disks, just had a question on that
      1) Why does Oracle access more than half of the voting disks?
      2) Why we keep odd number of voting disks?
      3) Does Oracle keep same information in all the voting disks ? is it like control or redofile multiplexing?
      4) Why we have only 2 ocr and not more than that?

      Regards
      ID
        • 1. Re: Does all voting disks in n node rac have same data
          Bjoern Rost
          I heard that Oracle/css access more than half of the voting disks, just had a question on that
          1) Why does Oracle access more than half of the voting disks?
          Usually (in a normal state of operation) all voting disks are accessed. But when there are issues with the heartbeat or interconnect communication between nodes, "pinging" these voting disks becomes important. The clusterware will use that information to determine if and which node needs to get fenced off (evicted) to guarantee consistency and to avoid a split brain situation. In that situation, one could generally say that a node needs to be able to access more than half of the voting disks to survive (it is actually a bit more complex than that)
          2) Why we keep odd number of voting disks?
          To guarantee an effective vote with a majority. Think of a two node cluster with two voting disks. A situation could occur in which the cluster is split and each node has access to one of the voting disks. How would the clusterware know which one is the "bad" node? This is a classic split-brai scenario and if both nodes keep accessing the same datafiles, corruption could occur. But if you have 3 voting disks and one node has access to two of them while the other one can only access one of them, the clusterware can assume that the one of the nodes is "more broken" than the other one.
          3) Does Oracle keep same information in all the voting disks ? is it like control or redofile multiplexing?
          The information should be the same across all voting disks.
          4) Why we have only 2 ocr and not more than that?
          I guess because someone decided that a mirror is enough and that we don't need to have (and keep) additional copies. Since we don't use those for voting, there does not have to be an odd number.

          cheers
          bjoern
          • 2. Re: Does all voting disks in n node rac have same data
            rarain
            Hi,

            1) Why does Oracle access more than half of the voting disks?

            To join the cluster greater than half number of voting disks must be accessible for the cluster node, Oracle made this restriction so that even in worst cases all nodes must have access on one common disk. Let's try to understand with simple classical example of two node cluster node1 and node2 with two voting disk vote1 and vote2 and assuming vote1 is accessible for node1 only and vote2 is accessible for node2 only, in this case if Oracle allow the node to join the cluster by passing this restriction then conflict will occur as both the node writing information to different voting disk but with this restriction Oracke make sure that one disk must be commonly accessible for all the nodes. For example in case of three voting disk at least two must be accessible, if two voting disks are accessible then one disk will be commonly accessible for all the nodes.

            2) Why we keep odd number of voting disks?

            I already answered this question indirectly with the answer of your first question. Greater than half number of voting disks must be accessible so either you configure three or next even number i.e. four but number of failover voting disk remains same. In case of three failure of one voting disk can be tolerated and same in case of four voting disks.

            3) Does Oracle keep same information in all the voting disks ? is it like control or redofile multiplexing?

            Yes, Clusterware maintains same information in all voting disks its just the multiplex copy.

            4) Why we have only 2 ocr and not more than that?

            We can configure upto five mirror OCR disk. Here is excrept of my ocr.loc file

            [root@host01 ~]# cat /etc/oracle/ocr.loc
            #Device/file getting replaced by device /dev/sdb13
            ocrconfig_loc=+DATA
            ocrmirrorconfig_loc=/dev/sdb15
            ocrconfig_loc3=/dev/sdb14
            ocrconfig_loc4=/dev/sdb13
            local_only=false[root@host01 ~]#
            [root@host01 ~]#
            [root@host01 ~]#
            [root@host01 ~]# ocrconfig -add /dev/sdb12
            [root@host01 ~]#
            [root@host01 ~]#
            [root@host01 ~]#
            [root@host01 ~]# ocrconfig -add /dev/sdb11
            PROT-27: Cannot configure another Oracle Cluster Registry location because the maximum number of Oracle Cluster Registry locations (5) have been configured

            Thanks
            • 3. Re: Does all voting disks in n node rac have same data
              1010036
              Thanks all, the solution has cleared some of my doubts , but still I have some queries

              2)Why we keep odd number of voting disks?

              you mentioned ,In case of three failure of one voting disk can be tolerated? that means cluster will run fine with 2 voting disks as well but there might be chances of 1 node accessing 1 voting disk and the other the second which can cause data corruption , so its always better to add 1 voting disk if you find only 2 running on your system? Correct me if I am wrong

              You mentioned " same in case of four" ? which is same as if 1 voting disk is corrupted then we have 3 , but again if we lost one more we can work with 2 right? Can you kindly explain this scenario?

              4) Why we have only 2 ocr and not more than that?
              You mean we can have max of 5 ocr , not more than that , so I can start with 2 and keep on adding till 5?

              Regards
              ID
              • 4. Re: Does all voting disks in n node rac have same data
                rarain
                Hi,

                As I mentioned greater than half number of voting disks must be accessible from each node then only node can join the cluster so either you configure three or four voting disk, the number of fail-over disk remains same i.e. one disk failure can be tolerated.

                Yes you can configure 5 multiplex copy of OCR disk.

                Thanks
                • 5. Re: Does all voting disks in n node rac have same data
                  1010036
                  Hi Rarain Thanks for the valuable information , i am done with Q 4 for number of OCR ?
                  But still not clear with Q2

                  Greater than half number of voting disks means in case of 3 voting disks 1 failure is tolerated, in case of 4 also 1 is tolerated but incase of 5 voting disks 2 failures are tolerated the third failure will boot the system ? Correct me iff i am wrong?

                  Regards
                  ID
                  • 6. Re: Does all voting disks in n node rac have same data
                    rarain
                    Hi,

                    Greater than half number of voting disks means in case of 3 voting disks 1 failure is tolerated, in case of 4 also 1 is tolerated but incase of 5 voting disks 2 failures are tolerated the third failure will boot the system ? Correct me iff i am wrong?

                    Correct...

                    Thanks
                    • 7. Re: Does all voting disks in n node rac have same data
                      Anju Garg
                      Please refer to following links :
                      http://oracleinaction.blogspot.in/2012/12/OCR.html
                      http://oracleinaction.blogspot.in/2012/12/votedisk.html

                      Hope it helps..
                      Regards
                      • 8. Re: Does all voting disks in n node rac have same data
                        1010036
                        Thanks all for the help