3 Replies Latest reply on Jun 3, 2019 7:58 AM by CHU TOULOUSE

    XFS architecture and configuration ?



      We want to create single instance databases Oracle EE and Oracle EE on a virtual machine Oracle Linux x86_64 bits, off course we want the best performance in i/o disk. 

      The XFS will be created on SAN Full Flash. 

      I want to split the database files on seven XFS like that:


      /oracle/oradata/<SID> <= data

      /oracle/oraindx/<SID> <= indexes

      /oracle/oradata/<SID>/redo01 <= redo group 1

      /oracle/oradata/<SID>/redo02 <= redo group 2

      /oracle/oradata/<SID>/ctrl01 <= controlfile 1

      /oracle/oradata/<SID>/ctrl02 <= controlfile 2

      /sv<SID>/archivelogs <= archive log


      Is this XFS architecture well define ?

      What are the good mount options, block sizes for each of theses XFS ? Is there a whitepaper about it ? Is the SAN disk blocksize important ? 

      Is there Oracle Linux 7 x86_64 systems configuration to improve i/o disk ? Is there a whitepaper about it ? 


      Thanks for your help and feedback.  



        • 1. Re: XFS architecture and configuration ?
          Avi Miller-Oracle

          I have moved this to the Database Installation space for better topic alignment.

          • 2. Re: XFS architecture and configuration ?

            You can partitioning for performance but this only makes sense if your underlying hardware is on a separate I/O bus. If all you do is to partition one and the same drive or device, or store data in different directories, it will do diddly-squat to improve performance. Given your layout, all you do is to store files in different directories.


            XFS, if I remember correctly, is supported with or later, but you'd have to check with Oracle support or find the appropriate MOS document. Keep in mind that you need RHCK or Oracle UEKR4 to meet compatibility specifications. You may find the following interesting:




            You certainly should not use any copy on write filesystem, such as Btrfs for database data files. You can probably also gain some performance by turning of filesystem journaling (at least that was the case in ext4). As far as I know, there are mount options for local storage that affect Oracle database. You specify the blocksize when you create the filesystem. XFS uses 4k by default. I can't imagine that changing this will make any difference, other than space efficiency. I'd go with the defaults. And btw, just in case, you don't need to worry about raw disk access - this is obsolete and Oracle database uses o_direct to access data files.


            However, ASM is the standard for Oracle Database. Why worry about XFS? With ASM, you don't have to worry about blocksize because database blocksize = ASM datafile blocksize. And btw, ASM does not interfere with physical I/O of the database accessing data. ASM (only) provides metadata to the database SGA, telling the database where to find data blocks.

            1 person found this helpful
            • 3. Re: XFS architecture and configuration ?
              CHU TOULOUSE

              Thank you Dude!

              I understand the i/o san bus layer is important...We only have one so ...

              OK. We are studying the project  to migrate Oracle EE 11g/12c databases from AIX 7.1 to Oracle Linux 7.6 x86_64.

              Maybe Oracle ASM will improve the i/o disk performance... But It will involve a lot of changes about the way we save our databases and I have not the knowledge of this technology.

              Our saves are done with Netbackup.  We supervise/aministrate with Oracle Cloud Control and sysload/EON.