Infrastructure at your Service

Jérôme Dubar

5 mistakes you should avoid with Oracle Database Appliance


It has been 5 years I’m working on ODA and I now have a quite good overview of the pros and the cons of such a solution. On one hand, ODA can greatly streamline your Oracle Database environment if you understand the purpose of this appliance. On the other hand, ODA can be your nightmare if you’re not using it as it supposed to be used. Let’s discover the common mistakes to avoid from the very beginning.

1st mistake – Consider the ODA as an appliance

Appliances are very popular these days, you need one feature, you buy a box that handles this feature and nothing else, you plug it and it works straight. Unfortunatly you cannot do that with an ODA. First of all, the level of embedded software is not clever enough to bundle all the features needed to simplify the DBA’s job. No doubt it will help you with faster deployment compared to home made environments, but all the tasks the DBA did before still exist. The appliance part is limited to basic operations. With an ODA, you will have to connect in terminal mode, with root, grid and oracle users to do a lot of operations. And you will need Linux skills to manage your ODA: it can be a problem if you are a Windows-only user. ODA provides a graphical interface, but it’s not something you will use very often. And there is still a lot of work for Oracle to gain the appliance moniker.

2nd mistake – Consider the ODA as a normal server

Second mistake is to consider the ODA as a normal server. Because it looks like it’s a normal server.

On the software side, if you ask a dba to connect to your Oracle environment on ODA, he probably won’t see that it’s actually an ODA, unless the server name contains the oda word 🙂 The only tool that differs from a normal server is the oakcli/odacli appliance manager, a command-line tool created to manage several features like database creation/deletion. What can be dangerous is that you will have system level access on the server, and all the advantages it comes with. But if you do some changes on your system, for example by installing a new package, manually changing the network configuration, tuning some Linux parameters, it can later avoid you to apply the next patch. The DBA should also keep off patching a database with classic patches available for Linux systems. Doing that will make your dbhome and related databases no more manageable by the appliance manager. Wait for the ODA-dedicated quaterly patch bundle if you want to patch.

On the hardware side, yes an ODA looks like a normal server, with free disk slots in the front. But you always have to order extension from the ODA catalog, and you cannot do what you want. You need to change disks for bigger ones? It’s not possible. You want to add 2 new disks? Not more possible, disks are sold as a 3-disk pack and are supposed to be installed together. ODAs have limitations. The small one cannot be extended at all. And the other ones support limited extensions (in number and in time).

Please keep your ODA away from Linux gurus and hardware geeks. And get stuck with recommended configurations.

3rd mistake – Compare the ODA hardware to other systems

When you consider ODA, it’s quite common to compare the hardware to what other brands can propose for the same amount of money. But it’s clearly not a good comparison. You’ll probably get more for your money from other brands. You should consider the ODA as a hardware and software bundle made to last more than a normal system. As an example, I deployed my first ODA X4-2 in May 2014, and the actual software bundle is still compatible with this ODA. Support for this ODA will end in February 2020, nearly 6 years of update for all the stack for a server that is able to run, and databases. Do the other brands propose that? I don’t think so.

What you cannot immediately realize is how fast the adoption of ODA can be. Most of the ODA projects I did are going on full production within 1 year, starting from initial ODA consideration. On a classic project, choosing the server/storage/software takes longer, deployment last longer because multiple people are involved, you sometimes get stucked with hardware/software compatibily problem, and you have no guarantee about the performance even if you choose the best components. ODA reduces the duration and the cost of a project for sure.

4th mistake – Buy only one ODA

If you consider just buying one ODA, you probably need to think twice. Unless you do not want to patch regularly, this is probably not a good solution. Patching is not a zero-downtime operation, and it’s not reversible. Even if ODA patch bundles simplify patching, it’s still a complex operation especially when the patch is updating the operating system and the Grid Infrastructure components. Remember that one of the big advantage of an ODA is the availabily of a new patch every 3 months to update all the stack: firmwares, bios, ILOM, operating system, grid infrastructure, oracle homes, oracle databases, … So if you want to secure the patching process, you’d better go for 2 ODAs, one for production databases and one for dev/test databases for example. And it makes no sense to move only a part of your databases on ODA, leaving the other databases on classic systems.

Another advantage of 2+ ODAs, if you’re lucky enough to use Enterprise Edition, is the free use of Data Guard (without Active mode – standby database will stay mounted only). Most often, thinking about ODA is also thinking about disaster recovery solutions. And both are better together.

5th mistake – Manage your Enterprise licenses as you always did

One of the key feature of the ODA is the ability to scale the Enterprise licenses, starting from 1 PROC on a single ODA (or 25 named users). 1 license is only 2 cores on ODA. Does it makes sense on this platform to have limited number of licenses? Answer is yes and yes. Oracle recommends at least one core per database, but it’s not a problem to deploy 5 or even 10 (small) databases with just 1 license, there is no limit for that. Appart from the CPU limitation (applying the license will limit the available cores), ODA has quite a big amount of RAM (please tune your SGA according to this) and fast I/O speed that makes reads on disks not so expensive. CPU utilisation will be optimized.

What I mean is that you probably need less licenses on ODA than you need on a normal system. You can better spread these licenses on more ODAs and/or decrease the number of licenses you need. ODA hardware is sometimes self-financed by the economy of licenses. Keep in mind that 1 Oracle Enterprise PROC license costs more than the medium-size ODA. And you can always increase the number of licenses if needed (on-demand capacity).

Buying ODA hardware can be cheaper than you thought.


ODA is a great piece of hardware and knowing what it is designed for and how it works will make you better manage your Oracle Database environment.


  • Hi Jérôme, very good advice. A must read before going to ODA.

  • Good article and very valid points!
    Also, disk speeds might not be the same between releases, as I found out when I replaced our X4-2 with an X5-2. We then had to order another storage expansion unit just to get the same IO performance as we had on the X4-2!
    After recent patching our X5-2 is now slower than our X4-2 ever was. Really disappointing.

  • Ivan Estigarribia Flores says:

    Excellent Article. A must read.

  • simon says:

    Thanks for sharing this post,
    is very helpful article.

  • Swadesh says:

    Very well done…
    in mistake -4 you mentioned about free use of data guard with out active mode. I guess, its pre-licensesd… we are planning to move our production load to the physical standby.
    Do we need separate license for open-readonly.
    Do you have any oracle note which can confirm that ODA is not licensed for active standby database

    • Jérôme Dubar says:

      Hi Swadesh, yes Active Guard is an option of Enterprise Edition. ODA has nothing to do with licenses. It’s a platform like another. You need to buy a license for Enterprise Edition, at least for 1 CPU or for 25 Named Users (if you only enable 2 cores on it). If you have 2 ODAs, you’ll also need at least 1 CPU license for Enterprise Edition on the second ODA, or 25 Named Users. Active Guard is an option, license is to buy accordingly to the Enterprise Licenses. For example, 2 ODAs = at least 2 CPU licenses (or 50 Named Users) = at least 2 CPU Active Guard Option or the option for 50 Named Users. Don’t forget that on ODA you have to disable the cores you won’t use, mainly because of the licenses. So consider the licenses on ODA the same as another x86_64 platform.

  • Chris Wheeler says:

    has anyone experienced an ODA randomly dropping network connections from anything outside of the subnet that it (ODA) is currently on?

Leave a Reply

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

Jérôme Dubar
Jérôme Dubar