Oracle Sun Will Try to Roadblock the Commodity Hardware Bandwagon
February 2, 2010
What’s the difference in cost between commodity hardware and branded hardware? The answer depends on how you count. I unearthed some old Google data years ago that suggested the Google could deliver orders of magnitude more performance with its home brew approach than it could achieve with name brand gear. The data were sufficiently obscure that my client at BearStearns was reluctant to include a 17X performance increase in one of our reports. I had enough Googley charts and tech articles to get the risk loving BearStearns’ crowd to go with a 4X benefit, but it was a tough discussion. My hunch is that Google won’t update or comment on how fast its home brew system goes. If these Google data were accurate, the implications for commodity hardware with the Googley fairy dust translates to a name brand stocked data center, running name brand gizmos would be more expensive than a Google equipped data center. Stated simply, if Google spends $1,000 dollars for a chunk of performance, a competitor would have to spend $4,000 or more to match Google’s performance. Big difference. Now you see why the 17X type of number made the BearStearns’ wizards nervous. $1,000 of Google gear translates to a whopping $17,000 of branded gear for comparable performance. With data centers hitting $650 million or more for outfits like Microsoft, the price tags to match Google become quite large.
When you say, RDBMS and performance, I see this technical diagram for addressing petascale data management challenges.
Now shift to the Oracle Sun deal. The Cnet write up “Oracle-Sun Versus Commodity Hardware” got my tiny goose brain turning in circles. Performance has been an issue with Oracle installations in my experience for a long time. The standard solution has been to throw hardware at the problem. Other RDBMS systems require the same remediation. Adding machines boosts the Oracle license revenue. For many years, the Fortune 1000 cheerfully pumped money into hardware and Oracle in order to keep response time within acceptable limits. Some companies such as the financial institution with which I worked a decade ago wanted me to find an alternative to throwing hardware at an Oracle system to speed up performance. I hooked the outfit up with a company called CrossZ, which has morphed into QueryObject. But most Oracle customers are happy to follow the recommendations of their Oracle DBA and the Oracle sales professionals.
The Cnet story included this interesting passage:
…with its newly acquired Sun hardware business, announced last week that it would go in the opposite direction and start selling direct in order to gain back the profit margin lost to VARs. As CNET’s Stephen Shankland wrote, Oracle is now a hardware company and needs to offset the fact that it owns a number of commodity products, including not just Sun servers but also MySQL and other pieces of software. By eliminating the middleman channel, Oracle can bump up margins. But it’s not clear that the market will be willing to pay a premium for Oracle-Sun products.
I disagree with Cnet’s belief that Oracle can make this variant of throwing hardware at a performance problem. Here’s why:
- The economic meltdown has reminded CFOs that the free spending days of yore are not appropriate for the present business climate. This means pushback from some Oracle clients who used to roll over like a dog under the ministrations of Caesar Millan, the dog whisperer.
- Hardware fixes to aging Oracle technology won’t do the job in the present world of big data. The Oracle database is not the right tool for big data. If it were, perhaps Google’s engineers, many of whom had some Sun experience on their bios, would have embraced Oracle. Google went a different direction, and I think it was a wise one as do the people who have suitcases of money from the Google home run.
- Innovators like Mark Logic are forcing Oracle to write quasi technical papers to point out that Mark Logic’s performance metrics are wrong. Nope. Oracle is wrong, and the problem is not Mark Logic. The problem is a variant of the Microsoft and other aging architectures. Just as the Model T cannot win a drag race against a hopped up Honda, Oracle cannot outperform a Mark Logic system. Old is to be respected. Old is not a solution for certain data management problems.
In short, Oracle purchased Sun for some good reasons. I am not sure hardware – software bundles and Sun servers offered to clients with tortoise like Oracle systems are among the best reasons. Just my opinion.
Stephen E Arnold, February 2, 2010
No one paid me to write this. A year ago, Mark Logic bought me a bagel. Since then, zip. I will report the food payment to the FDA in the morning.
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
