This research has been significantly extended in two more recent professional alerts (Spring 2014). These are:
- "Duplicating Public Cloud Economics for Oracle Database Infrastructure";
- "Duplicating Public Cloud Economics for Oracle DBAs".
Please see Comparing Oracle Exadata Storage with Best-of-Breed Arrays for more detail on technical assumptions.
Oracle Exadata, based on Sun hardware, was announced in 2010 and has been aggressively sold by Oracle since. The value proposition to Oracle developers is clear – a single appliance with a single part number (SKU) that would allow Oracle databases to be managed by the DBA and perform without operational staff. A development manager’s dream, it would appear.
Wikibon has looked at this value proposition in detail two years later and has concluded that it is a very expensive dream in almost all situations. A conservative case study in Figure 1 below shows that the Oracle Exadata is 83% more expensive than alternatives from competing vendors over three years, with an additional cost of about $3 million dollars.
In addition, Wikibon has concluded that Exadata is not a best-of-breed solution for very high availability and/or very high performance mission-critical OLTP. The Oracle Exadata servers are not best of breed, and the storage subsystem is far below the functionality required. Tier 1 storage arrays such as the EMC VMAX and other Tier 1 platforms offer significantly high performance and availability options at significantly lower cost, together with the ability to share storage across database instances and across the data center as a whole.
Special Exadata Oracle features such as IO resource manager, Smartscan, and Hybrid Columnar Compression do not help in the vast majority of OLTP systems – in most cases they get in the way. Figure 2 shows that more flexible general purpose appliances or best-of-breed servers and storage are more cost-effective and higher function solutions.
For data warehousing workloads, the special Exadata features can help to reduce the elapsed time of jobs when non-indexed full table scans are key. In most practical cases, similar performance can be achieved using indexing (so the additional cost of indexing was included in the business case results shown in Figure 1). Figure 3 shows that Oracle Exadata is more expensive than the alternatives and does not compete well against shared-nothing data warehousing solutions (e.g., EMC Greenplum, HP Vertica, IBM Netezza, etc.) when high-end performance is required.
Most data centers, need separate OLTP, data warehousing and development environments, as well as remote business continuance. Oracle Exadata is a closed environment, and it is difficult and time consuming to move data between an Exadata infrastructure and alternative systems. As a result Exadata customers have found that they need to put all the Oracle database environment on Exadata appliances. This is a great lock-in for Oracle, but adds to IT expenses significantly. In the case study shown in Figure 1 above and Table 2 below, the OLTP system was installed in Year 1, and the additional data warehousing and development systems in Year 2.
Wikibon concludes that Exadata as a closed system is about twice as expensive as alternative approaches. In some cases, this may be justified if access to Oracle expertise is critical for the development of a project. In most, the savings resulting from using lower cost infrastructure could be applied in part to purchasing database services to accelerate the project.
Wikibon has also analysed the Oracle Exadata and Exalogic sales, and concluded that the market for direct sales just to development departments is limited.
Action Item: CIOs, CFOs & CTOs should understand that the value of Exadata is limited to the business relationship between development staff and Oracle. The Exadata hardware is relatively costly and has performance limitations compared with marketplace alternatives. CIOs & CFOs should require a full business case before considering signing off on Exadata deployments, and in particular ask how the savings from using a general purpose and best-of-breed solution could be used to help reduce implementation time and time to value.
From Oracle’s perspective, their approach to generating hardware sales risks long-term dilution of its well-earned, trusted relationship with its software customers.
Footnotes:
Methodology:
The methodology used in this study comprised six major components, including:
- In-depth interviews with more than 10 Wikibon practitioners that were either Exadata users, users of general purpose array hardware or users of converged infrastructure products. Several of the interviewees had experience with each.
- Interviews with six expert database consultants that had experience with hardware and software products from Oracle Exadata and competitive offerings
- The development of a return on asset (ROA) model specific to Oracle Exadata and general purpose bespoke arrays. An ROA model provides a way of assessing the profile of an installation and understanding the financial impact over a period of time. Importantly, unlike many ROI models, an ROA model does not assume brand new technology is installed, rather it uses a mix of new and older equipment evaluating the impact on the overall base as a whole. More information on the ROA methodology can be found here.
- A detailed technical comparison performed by Wikibon analysts based on publicly available information - Please see Comparing Oracle Exadata Storage with Best-of-Breed Arrays for more detail on technical assumptions.
- Pricing analysis based on data from Oracle's Web site and data from general purpose single SKU appliances in the market.
- Assumptions that fed both the ROA model and pricing analysis as detailed below.
Notes on Table 1:
- Year 1 is the introduction of Oracle Exadata Appliance or general purpose appliance. The reference model for the appliance is the VCE.
- The Oracle Exadata is over configured for OLTP, with performance disks only and additional hardware that is not required. The alternative general purpose appliance come in at 65% of the Exadata hardware price, because it can do the same work with 65% fewer resources.
- The original Exadata is sold with the ability to run OLTP, data warehousing and development/test. In year 2 this vision was abandoned for performance reasons, and separate data warehousing and development systems installed. The price of these two systems reflects a 30% reduction in year-on-year IT costs.
- Year 2 for the general purpose appliance sees the introduction of another general purpose appliance for data warehousing, and the use of existing data center assets. The cost of existing assets is shown as depreciation charges and maintenance. The use of virtualization for test and development also increase utilization and development productivity.
- At the end of the three year business case, there is a resale value of two Exadata systems in the case of the Oracle scenario, and two appliances in the case of the general purpose solution. The general purpose appliance will have a higher % resale value.
- The general purpose appliance does not include Oracle Databases as part of the stack. Additional effort is required to install and maintain the Oracle databases, and ensure that they are kept in line with the general appliance stack.
- The Oracle Exadata has additional function to allow faster non-indexed full table scans. Additional development resources have been allocated to allow indexing of the tables.
- The cost of the additional Oracle licenses is assumed to the the same cost as the first Exadata ($1m). The Oracle licenses are assumed to be relative to the amount of hardware required (65% of the Exadata).
- Hardware and software maintenance is assumed to be 18% of the list purchase price of the equipment installed.
- The Net Present Cost uses a discount rate of 5% to normalize the time value of money.
The table concludes that the Exadata appliance is 83% more expensive that the general purpose appliance. The savings from the general purpose appliance is about $3 million.
Wikibon concludes that these savings could be used to buy additional services to help implement the Oracle system more quickly, and achieve the required business benefits earlier.