13 Responses

  1. Simon Williams
    Simon Williams at |

    Even faster when virtualised on Fusion-io. Fusion-io is great for consolidation and speed. We ran Oracle that was on 2 racks natively on 2 x 1 RU servers and it was 6x faster.

  2. Richard Houghton (@S
    Richard Houghton (@S at |

    Hey Michael, another nice post and good timing for me 🙂 In your experience of moving Oracle off UNIX onto Linux, what Linux are customers choosing? You mention Suse (nice, didn't know about entitled use) but are people using this over say RHEL or Oracle Linux?

    1. @vcdxnz001
      @vcdxnz001 at |

      Hi Richard, most people don't know about the SLES entitlement. The most popular destination OS by far is RHEL.

    2. Larry Timmins
      Larry Timmins at |

      Hello Richard,

      Since July 10th, I recently benefited from discussions with Oracle experts in professional services organizations for (possibly) unbiased view of what Linux to deploy.

      First a Summary of Choices:


      (1) Redhat RHEL 6 – has paravirtualization, same kernel as Oracle Linux 6, no ASMLib. But RAC and ASM are supported by Oracle on RHEL 6.

      (2) Oracle Linux 6 – stripped RHEL kernel (no PV drivers) supposedly to benefit Oracle VM benchmarks over virtualization competitors to Oracle, but I haven't seen any follow-up / results to back that up. Obviously, ASMlib is part of UEK 6 with Oracle Linux 6

      (3) Hybrid: Pay for both Oracle Linux support and Red Hat Linux support and you can combine a Oracle UEK6 kernel with ASMLib with your RHEL 6 environment

      (4) RHEL 5.4 with ASMlib — up to recently – deployments of Oracle appeared to be split between RHEL 5.4 and 5.6.

      (5) RHEL 5.6 with ASMlib – see above comment on choice (4)

      (6) RHEL 5.7 with ASMlib – latest RHEL with "no problems with ASMlib"

      (7) RHEL 5.8 with ASMlib — however — some reports of "ASM issues"

      Decisions, decisions:


      From multiple Oracle consulting sources, many large enterprise accounts are not moving to Oracle Linux 6 or RHEL 6 just yet. I actually received feedback that Oracle's marketing choice may backfire against a migration of Oracel ASM accounts to Oracle.

      Many shops remain on RHEL 5.x and have chosen to move to a recommendation of RedHat 5.7 with ASMlib for new deployments.



      In the VMware space, if relevant to your shop, the recommendation is overwhelmingly to move to latest vSphere 5.0-Update1 or later to gain inherient hypervisor performance although those that stayed with VMware vSphere ESXi vSphere 4.1 Update 2 / latest patch have no complaints.



      Another issue that plagued me was whether to even consider using RDM (raw device mapping) or VMFS. With physical servers, the decision was easy – map your SAN LUN to a physical machine running a native OS and its dedicated. With VMware, you can have multiple VMDKs on a single LUN. That gets more complicated for Oracle storage layouts. The trick is to know where sequential I/O will occur and then dedicated a single VMDK for Oracle to a single LUN. That gives your the traditional benefits of RDM (other than ease of converting VM to a physical host) and you overwhelmingly benefit from Storage vMotion, etc which I consider more important.

      The reason I bring that up is because the same kernel that Redhat and Oracle Linux use no longer supports /dev/raw anything and you either play with "udev" etc to have your drives mounted , etc.

      I think the combination of proper storage planning for Oracle VMs; separation of Redo logs, Ora data, Flash recovery area, and even Temp space; and

      using a dedicated LUN where sequential I/O is a performance requirement will give you the most that Linux, your server, storage network and SAN can deliver.

      More than Linux question


      I recently found this blogger last week and there is exceptional information here on a lot of the blogs that confirmed or gave examples to help my decision making for our Oracle environment decisions — Oracle is the last non virtual environment in our shop.

      Not sure about tiered storage pools yet, but we're going to start with testing enhancing native array cache with SSD (200GB to 400GB) since most of our data warehouse I/O operations are below 128KB and should fit the extended cache. That will allow our many small I/O operations from OLTP to fit into the native array cache (64GB or less) and we should see top performance on our Oracle on a Linux VM servers.

      All the best,

      Larry T.

      1. @vcdxnz001
        @vcdxnz001 at |

        Hi Larry,

        Thanks for the great comment. There is some really useful info there. One thing you should seriously consider is adding FusionIO and IOTurbine into the mix. It'll depend on the version of hypervisor you're running which version of IOTurbine etc you need. I'm currently testing Beta that works with Linux and vSphere 5. I will be posting a separate article about this. On a pure IO side using the FisionIO card as a datastore I was able to get over 100K 8KB IO's at 2ms latency 100% random either read or write. So that's not bad especially when comparing the price of even the 1.2TB IODrive2 card I've got vs a SAN upgrade. Using IOTurbine the VM's can still be vMotioned, it just offloads read IO caching from the SAN / OS to the FusionIO card, writes are passed through to storage to ensure data integrity, but still retained in cache. It might not be for you, but it's worth a look and checking it out at least.

        On the sequential vs random IO front and whether or not to use a single datastore for a single VMDK, it'll come down to how best your storage handles the IO. Many storage arrays are optimized for random IO and not sequential. So depending on your storage it might pay to have multiple VMDK's on the same datastore in that case. You can also get better IO and storage utilization by grouping low IO consumers with higher IO consumers (OS VMDK with Redo VMDK like I did in an example). You also need to watch the queue depths and how much each device can handle. If you've set your queue depth to 64 and you've only got a single VMDK on a datastore then it's not going to use the full queue depth that is available. But a lot of care is needed as you don't want to overload your storage processor queues. The database is an extension of storage so an in depth understanding of your storage environment is required in order to get the best possible performance and assure SLA's are met.

        I haven't yet posted anything on ASM, but I hope to at some stage. I do have some slides on it at my VMworld 2012 presentation Virtualizing Oracle: An Architectural and Performance Deep Dive – APP-BCA1624.

  3. Timbo
    Timbo at |

    Hey Mike. Regarding your first point. Here is a good Forrester paper on the Total Economic Impact of virtualizing Oracle on vSphere. http://www.vmware.com/files/pdf/solutions/total-e

  4. claudiog64
    claudiog64 at |

    I would say "For Those About to RAC We Salute You" 😀

    Nice post.

  5. 10 Reasons to Migrate Oracle Databases from Traditional Unix to Linux on vSphere | Big Data your head in the clouds | Scoop.it

    […] This article provides an overview of what I believe are 10 good reasons why you should seriously consider migrating Oracle Databases off traditional Unix platforms to VMware vSphere based on my exp…  […]

  6. Fight the FUD – Oracle Licensing and Support on VMware vSphere « Long White Virtual Clouds

    […] of the migration project instead. I have outlined in a previous article what I think are the Top 10 Reasons to Migrate Oracle Databases from Traditional Unix to Linux on vSphere. License Maintenance costs are not the only cost savings by switching to x86-64 from traditional […]

  7. Planning to Migrate Oracle from Unix to Linux on vSphere « Long White Virtual Clouds

    […] Why Linux on vSphere? (Repeated from my article 10 Reasons to Migrate Oracle Databases from Traditional Unix to Linux on vSphere) […]

  8. mark
    mark at |

    have you ever used ioturbine and asm? do they work?

    1. Simon Williams
      Simon Williams at |

      Mark, yes. ioTurbine supports RHEL & SUSE Linux VMs. Happy to provide eval kit if you're in ANZ. For more info, take a look at: http://www.fusionio.com/press-releases/fusion-io-

    2. @vcdxnz001
      @vcdxnz001 at |

      Hi Mark, As Simon said it works. I haven't yet tested it myself but it's on my list to do in the near future and report on the performance results. So keep an eye out for an article about it.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.