Oracle 10g v. SQL Server Because you have a right to my opinion

Total Cost of Administration:
Oracle 10g vs. SQL Server

In a classic Apples vs. Oranges comparison, ComputerWorld tries to sensationalize a comparison of Oracle 10g vs. SQL Server administration costs.  It's like comparing a Volkswagen to a Ferrari.  They are both vehicles, but the similarities stop there.

So, what do we make of this "study"?  It starts by correctly noting that human costs comprise the majority of IT budget, and that reducing the number expensive DBA's is an area of opportunity for the IT manager to save money. 

Now I ask you, how do you expect a DBA to react when reading an article that is focused on reducing the DBA staff size?  I know how managers think, and this article could wind-up in the wrong hands (envision the pointy-haired boss in "Dilbert").

Over the years, I've read enough jaded studies to suspect when a study may be biased, so I did some checking.  The company who did the study (Alinean Inc.), has a web page on their site that is hosted by Microsoft (Migration Calculator), and the page says "Microsoft Partner Program", indicating that the study was done by a Microsoft partner.  At the bottom of that page, also note "This site hosted for Microsoft by Alinean, Inc." 

The comparison's conclusions

Anytime I see anything that smells of hyperbole in a study, I become concerned about credibility and possible bias.  The quote below caught my eye because it reminded me of an infomercial I saw last night that used exactly the same phrase "That's a savings of more than . . . ":

"On average, the annual cost for administration is $2,847 per year per database for Microsoft SQL Server 2005 and $10,206 per year per database for Oracle 10g.

That's a savings of more than 350% in annual costs per database for the Microsoft platform."

Among the conclusions in this article, we see this claim that there is a large difference in the amount effort to install SQL Server vs. Oracle 10g:

"The results revealed that, overall, Microsoft SQL Server 2005 required significantly less effort to install and maintain than Oracle Database 10g."

 I've heard rumors of an Oracle demo where a 12 year-old girl installs Oracle 10g, and I'll bet that any smart six year-old can install SQL Server.  If you have a six-year old at-home who wants to install Oracle, please send me an e-mail at info@remote-dba.net. 

But wait, there's more! 

Here we see a suggestion that Oracle databases are somehow "more work" than SQL Server databases, in this case, by a factor of 3:1:

"Study participants reported that on average a database administrator could manage more than 30 Microsoft SQL Server 2005 databases, while Oracle 10g implementations required one DBA per 10 databases."

I can't speak for the average DBA, but I've personally managed over 50 Oracle databases, and with the proper automation in-place, a good DBA can manage hundreds of instances.  Also, note the categorization of "large volumes of users" in the methodology section of the research study. 

Oracle guru Jeff Hunter (there are two well-known Jeff Hunters in Oracle), suggests that this wad-of-phlegm study is largely meaningless because Linux is far more powerful than Windows:

"I can't believe people are actually paid to come up with these numbers. Lets see an apples to apples comparison.

For example, put Oracle and MS SQL on the same hardware on Windows XP (yes, I'm going against my "Never Windoz" philosophy, but last I checked, MS SQL wasn't available on Linux). "

Buried in the second page, we see a disclaimer to their 350% savings claim, and the truth peeks out:

"So as database size, complexity and workload grow, the results begin to even out. And for the largest databases, Oracle appears to be slightly less expensive to administer."

The conclusion also sets tiny cut-offs for a database of "reasonable size", perhaps not understanding that SQL Server is used for data volumes that could be kept inside a spreadsheet.  In plain English, SQL Server counts itty-bitty files as databases, perhaps unfairly comparing the TCA of a 5k databases with a 5g database, without considering that the Oracle databases is several orders-of-magnitude larger:

"When looking at larger databases where overall size and scalability are the critical factors, Oracle Database 10g appears to take the lower TCA prize.

But where databases are of reasonable size and/or where there are more databases and database users to manage, not unexpectedly Microsoft SQL Server 2005 has the TCA advantage."

When I was in college statistics we were required to read the great book "How to lie with statistics", and I have always remembered the lessons about how you can twist your dataset to arrive at any conclusion you desire.  Plus, in a trick worthy of Machiavelli, you can claim that the numbers "prove" the assertion with a confidence interval of two standard deviations!  For an example of broad numbers, in the study whitepaper we see this estimation of the salaries of SQL Server vs. Oracle DBA staff:

"For Microsoft DBAs the calculation was: $82,500 - (11 – 6.1 years)*4% or $82,500 - $16,170 = $66,330

For Oracle DBAs the calculation was: $88,000 – (11 - 7.2 years)*4% or $88,000 - $13,376 = $74,624"

We can only hope that managers don't take this article too seriously.  At the end of the day, it's robustness and reliability that win the day, not the total cost of administration.