Is Intel the right platform to run Oracle

It doesn't have to be Ford v Chevy

There are several competitive adages like Ford vs Chevy, John Deere vs Case International, Honda vs Toyota, Mac vs PC, Coke vs Pepsi and so on. Please share some of your examples in the comments. Some of these comparisons are tongue in cheek while some people take it quite seriously.

Several years ago, I had a client running their multi-billion dollar business with a small mainframe and a modest IBM i system. The mainframe hosted the financials while the IBM i system, which was configured to use 13 IBM Power cores ran the manufacturing, inventory, and transportation. Their CIO had been convinced a company their size should be running a legitimate ERP solution from SAP, Oracle, and Infor. They chose Oracle EBS (E-Business Suite) running on Intel servers with VMware and Linux.

The consulting effort began and at some point along the journey they had their first Oracle EBS workload running in Production. I was told this represented roughly 25% of the architected end-state. To put things into perspective, this 25% had accumulated 1,000 Oracle licenses. That is 1,000 Oracle Enterprise Edition database licenses. Those familiar with Oracle licensing, as it was the case here, they used a processor/core metric as defined by the Oracle Processor Core Factor Table. For Intel systems, you took the total number of cores in each running Oracle products times 0.5. Since this is 1/2, we know if there were 1,000 Oracle licenses there were 2,000 cores. With only 25% in, the CapEx and OpEx costs were blowing up. How much would it cost to finish the remaining 75%? That was a question their parent company was asking. They were granted some leeway to continue on but with strict guidance to get costs under control.

I had been pursuing this client to initially run the Oracle workload with AIX on IBM Power but the client resisted calling IBM Power a legacy platform while Intel+VMware+Linux was the future of computing. I didn't get the initial call but after their financial review, I was asked to participate in the next phase. There were 2 Intel options and 1 IBM Power solution being considered. The first Intel option was a major converged infrastructure (CI) solution while the 2nd Intel solution was an appliance. IBM Power, being a general-purpose computer with tremendous granularity and workload segregation, was (is) ideally suited for any enterprise software licensed by processor/core.

The solution was a 2-site solution. Setting aside the Intel appliance option as it had its own deficiencies it also makes it more difficult to compare apples to apples due to how the vendor shifts cost around the total solution. Focusing on the CI and IBM Power solution. The client provided us both the sizing data for every workload and location we were to consider.

The CI vendor did their sizing and I did the sizing for IBM Power. Looking at just 1 location of the 2, just for the Oracle EE DB licenses which I'll refer to as Category 1 as there were 2 other major categories for both locations. Category 2 was Oracle EE DB + Oracle RAC while Category 3 was Oracle WebLogic. Back to Location 1 for Category 1. The CI vendor stated the required number of cores would be 300 Intel cores. This translates to 150 Oracle EE DB licenses. The 300 core number is the sum of all the individual servers. If a DB required 5 cores, the server might be sized to have 8 cores and so on. If you totaled the required cores, it was 184 or 92 Oracle licenses. This isn't possible as it assumes the server is running at 100% utilization with no room for growth not to mention there is not a 5 core server available so there would always be some degree of "rounding up".

I did the Power sizing. For Location 1, Category 1 the requirement was 28 cores. The licensing factor per the processor core factor table for IBM Power is 1.0 core = 1 Oracle license, thus 28 Oracle licenses.

It should be noted how the costs were calculated. What you see below is the List Price x 75% discount. This amount is the license cost. Because Oracle charges software maintenance for the 1st year, take the license cost x 22%.

Ex: $47,500 x 25% = $11,875 per Oracle EE license. $11,875 x 1.22 = $14,487.50 is Oracle license + 1st year software maintenance cost. $14,487.50 x 28 = $405,650.

Some immediate observations. The Power solution is not 10X higher performing but it is 10X more efficient (based on this sizing). IBM Power is not 6X higher-performing if we were to use the 184 cores vs 28 cores, this is a combination of performance as well as efficiency obtained from the built-in Power Hypervisor with its ability to segregate VM's. With granularity to allocate physical cores ranging from .05 of a physical core to any increment up to the max number of cores in the server. Thus, VM1 might require 0.6 cores. VM2 might require 1.2 cores and so on. The sum of these equals 30 cores. Note this contrast. 300 vs 28, 150 vs 28, 184 vs 28 and 92 vs 28.

By virtually every metric, IBM Power was (is) the optimal platform for this (every) Oracle workload with far fewer servers, all fully-virtualized, far fewer cores and far fewer Oracle licenses. Contrast this with the Intel-based CI solution, which did not use VMware for any of the database workloads (Oracle EE + Oracle RAC) due to worries of a prolonged Oracle audit and legal battle on whose interpretation is correct regarding licensing Oracle in a VMware environment. With no virtualization, they require many more servers, many more cores, and many more Oracle licenses. Though not shown here, the cost of infrastructure for either solution was very similar. That is not necessarily the situation today given the incredible performance advantage IBM POWER9 has over Intel today not to mention the efficiency obtained from the Power Hypervisor.

As seen in the image below, we have not discussed the sizing and cost for Location 1 & 2 for Categories 2 and 3. This image shows the sum of both locations for all 3 categories using the same discount and method to calculate as shown above.

The 5 year TCO which includes the initial Oracle license cost plus years 1-5 of Oracle software maintenance increases the delta even more between the two Intel solutions and IBM Power. I encourage you to do the math.

If you are already running Oracle on Intel, that license cost is a sunk cost as you likely purchased perpetual licenses. Deploying new Oracle environments shows a clear advantage to IBM Power. Changing (migrating) platforms running Oracle workloads is a well-understood process that is very mature. The experience, tools, time and cost are likely nowhere close to what you think it is. Anecdotally, I've had many C-levels discussions regarding their Oracle environments. They often state their code is highly custom making it far too expensive to switch. These beliefs are often assumptions not based on any discussions with migration specialists. This issue does not have to be a Ford vs Chevy debate. Those are fun to debate at dinner, the gym or the bar. However, business leaders who want to control CapEx and OpEx have options. Choosing the right tool for the job is the difference from using a wrench that fits the nut perfectly vs an adjustable wrench which often is not the right tool possibly causing more problems and costs down the road.

Contact Us

To learn about our core business offerings, visit Clear Technologies, Visual Storage Intelligence, and our AI practice; Clear Intelligence.

About the author: Brett Murphy is a Client Executive and Executive Architect at Clear Technologies. With over 25 years of industry experience, Brett is a Technologist, Architect, Speaker, Author and IBM Champion specializing in enterprise workloads such as SAP ECC, SAP HANA, Oracle DB, Oracle Financials, Oracle EBS, PeopleSoft, JD Edwards and more. Follow Brett on twitter @PowerMan4EVR and his blog https://powertheenterprise.wordpress.com.