38 Comments

Summary:

There’s a reason Larry Ellison called cloud computing “nonsense” and why he still won’t permit Amazon-style metered pricing for Oracle’s mainstream database and middleware. George Gilbert lets us in on the company’s expensive secret.

Gilbert_CloudComputingNonsense_secret_image

There’s a reason Larry Ellison called cloud computing “nonsense” in 2009 and why he still won’t permit Amazon-style metered pricing for Oracle’s  mainstream database and middleware. A traditional 11g database license that today costs $2.8 million up front would cost less than $9 per hour using Oracle’s mySQL on Amazon. (Keep reading to see why this apples-to-oranges comparison is valid.)

We’ve seen a similar scenario play out before — back when IBM mainframes ran mission-critical applications on legacy databases. IBM actually pioneered relational databases, but it was conflicted about selling the lower-priced, lower-margin servers needed to run them.

These servers had the price-to-performance ratio customers needed for the performance-hungry RDBs. So a new generation of infrastructure vendors — led by DEC, HP, Sun, Microsoft and Oracle — disrupted the old IBM platform. Just like IBM, Oracle has the technical wherewithal to compete with the new databases that are powering cloud-based applications, but they’re conflicted about how to handle metered pricing in these environments.

Metered pricing disrupts old business models

We are in the midst of at least two technology disruptions. But as Clayton Christensen described in “The Innovator’s Dilemma,” disruptions are more often about addressing the needs of “un-served” and “over-served” customers than they are about revolutionary technologies. Web 2.0 apps are a perfect example of customers over-served by Oracle’s enterprise database. During the dot-com bubble, Oracle’s enterprise database ran on big Sun and EMC boxes and powered both Web and SAP-class applications. Since then, however, untold numbers of Web apps moved to the low-end and less expensive mySQL as part of a migration to the LAMP stack.

The second disruption is reaching un-served customers in social media and other new markets who are building big data applications with new levels of data volume, variety and velocity. These customers are often using the SMAQ stack or the still-emerging class of NoSQL or NewSQL databases.

Amazon enabled both of these disruptions by offering two critical features. They enabled on-demand delivery with elastic capacity, using hourly metered pricing and the ability to automate complete control of the remote hardware infrastructure, as Scalr CEO Sebastian Stadil recently explained to me.  (Scalr manages applications for thousands of customers on Amazon and other service providers.)

Traditionally, the most challenging disruptive innovations force incumbent vendors to change their business models. Oracle clearly has the technical wherewithal to build databases that meet the needs of Web 2.0 and big data applications. But changing the resources, processes and values that underpin its business model in order to support metered pricing will be immensely challenging.

A closer look at traditional and metered prices

Even if Amazon fully supported a typical Oracle configuration, Oracle’s bread-and-butter enterprise edition database in a two-node cluster with RAC and caching on Amazon’s largest EC2 databases would cost more than $900,000 in upfront licensing fees. But according to Scalr’s data, a typical application runs most of the time at only 40 percent of its peak capacity. Since Oracle requires buying licenses for peak capacity, a $900,000 cluster would be upsized to $2.3 million. With the obligatory pre-payment of 12 months’ maintenance, the initial commitment totals $2.8 million.

Compare that to a baseline cost of $5.20 per hour for the same configuration of mirrored mySQL database servers. Peak demand would top out at $12, but it would hit that only periodically, such as during a holiday shopping surge. Scalr’s data also indicates that when capacity is averaged out between peaks and a 40-percent baseline, it comes out to 60 percent of peak capacity. So with metered pricing, if we start from $12 an hour for peak capacity, that totals an average cost of $8.70 an hour, or $19,000 per quarter or $76,000 per year.

The traditional, upfront model has additional charges. According to Michael Crandell, CEO of RightScale (a company that has launched more than 3.5 million servers for customers on Amazon and other service providers), managing the full application lifecycle may require additional licenses. Server licenses for use in quality assurance, load testing, staging and standbys for failover might not be included in the production license. Add those to the $2.8 million.

How metered pricing disrupts Oracle’s business

The most important number to traditional enterprise software companies is their total revenue during the quarter they make a sale. The bigger that number, the more profitable the company looks after subtracting the heavy, and relatively fixed, upfront expenses for sales and marketing and R&D. Oracle today recognizes immediately $2.3 million after subtracting the 12-month maintenance subscription of $0.5 million from the $2.8 million total.

At the risk of greatly over-simplifying its published income statement, let’s say that Oracle would then subtract one-third, or $750,000-plus, for sales and marketing and R&D. (In reality, the sales and marketing expense for license revenue is actually higher, because follow-on maintenance services take little effort to sell, and dwarf the license revenue). The remaining $1.5 million would be Oracle’s profit margin for the quarterly reporting period when they made the sale. Now subtract that same $750,000 in expenses from the $19,000 in quarterly revenue under metered pricing. That comes to a loss of $730,000 for the quarter in which the sale is made.

In fairness, these metered revenue streams would add up on top of each other. But the point is that the transition to metered pricing would dramatically erode Oracle’s revenue and 45 percent of its operating profit margins. That is why Oracle is resisting metered pricing.

The bottom line

Contrary to Oracle’s claims, neither its Exadata database machine nor its database appliance is a cloud strategy. Both strategies base their pricing on peak capacity, not on elastic metering. Furthermore, discrete hardware is the opposite of cloud infrastructure, which enables near-infinite capacity on demand. Like IBM during the client-server transition, Oracle has the technology to address customer demand. It is just conflicted about the business implications of cloud computing’s metered pricing.

George Gilbert is a co-founder at TechAlpha Partners, a consultancy that works with vendors serving the enterprise market, startups and institutional investors on issues of business strategy and product management and marketing. Previously, Gilbert was the lead enterprise software analyst for Credit Suisse First Boston, a leading investment bank in the technology sector.

Image courtesy of Flickr user stevendepolo.

For more on how big data is impacting the tech industry, be sure to check out GigaOM’s Structure:Data Conference in New York City on March 21 and 22.

  1. Nice post ! Conclusion: go short on Oracle ;)

    Share
    1. George Gilbert Saturday, March 10, 2012

      Steve – there’s an old saying that one can go broke waiting for a short position to work, even if you really are right. this will probably take some time to play out. a good rule of thumb of these types of transitions is that it takes longer to happen than expected but then happens faster than expected when it finally does happen

      Share
  2. Good take on metered pricing in general but IMHO a couple of big flaws
    a) No one pays full price on an Oracle db – how would the avg discount tie in with the
    b) Most importantly – a mirrored MySQL configuration is *not* equivalent to an Oracle RAC configuration from many angles (performance, scalability, features). If it was, what prevents users from deploying MySQL in production today – metered pricing or not? after all MySQL is free. If MySQL is a perfect substitute in the cloud its a perfect substitute on premise and the pricing argument above falls apart

    Share
    1. George Gilbert Saturday, March 10, 2012

      You are right that most customers don’t pay full price on Oracle enterprise edition. But about 10 years ago they started standardizing the discounts worldwide and exceptions have to be approved by the CFO or someone close to that level. So yes, the “full” price would be lower, but it’s still orders of magnitude higher than metered pricing.

      You bring up a more substantial point with the comparison to mySQL. It has become a widespread substitute for the Oracle/Sun/EMC stack of the dot.com bubble in Web sites from the largest to the smallest. A common configuration, not quite the same as Oracle, is to deploy a “master” node that runs on the maximum size EC2 instance – since mySQL doesn’t handle write replicas in a clustered configuration. To scale the reads, many sites deploy read-replicas (slaves). More recently, many have taken to deploying caches like Memcached to scale the reads. The reason mySQL hasn’t substituted for Oracle’s enterprise database in traditional, on-premise, SAP-like applications is that the above configuration would break them. Traditional applications assume they don’t have to know anything about the physical layout of the data, which Oracle provides. So you can’t really slide mySQL underneath these applications. It’s possible this will change over time as the applications get modernized, but i wouldn’t hold my breath.

      Share
      1. So lets compare apples to apples. If you compare pricing for Oracle (metered vs not) does it affect Oracle’s economics substantially to offer a 40% discount? Thats part 1. As for MySQL, thats my point. Customers who can use MySQL on premise already do (or will soon). They pay 0$ (unless they want support from Oracle) and a lot of the web companies dont pay any support fees to Oracle. Also, many of these customers have moved on to the NoSQL stuff for things they used to put on MySQL. The customers who *cant* use MySQL have to stay with Oracle. We seem to be in agreement up to this point. Now, the leap the article makes (and that I disagree with) is using the *metered* price of MySQL to the on-premise cost of an Oracle db is sort of meaningless, isnt it? Metering wont disrupt that model . MySQL (or an equivalent metered competitor) getting *good enough * will disrupt that.

        As for the metering impact on financial profitability – thats a different point. I would argue that metering better be financially profitable. Are Oracle’s expenses (on R&D for e.g.) out of line from what other metered vendors (say CRM) spend? If not, then the transition to metered pricing better cover those costs or these companies wont exist. Now its possible to have a Craigslist like disruption here. A competitor survives on significantly less revenue because their cost structure is so radically different. This shrinks the size of the pie but increases that competitors net income. Do we see that happening here?

        Share
      2. George Gilbert Sunday, March 11, 2012

        Anon – See below for a reply to your reply. Can’t do 3 levels.

        Share
  3. Dominic Delmolino Saturday, March 10, 2012

    Good post — by why single out Oracle? IBM, MSFT, SAP — they all have the same problem. And actually, you can get meter-priced Oracle through the Amazon RDS service, another point missed in the article. I agree meter-based pricing is a disrupting concept, but it applies to the whole enterprise software market, not just Oracle.

    Share
    1. George Gilbert Saturday, March 10, 2012

      Dominic – you asked the question i was waiting for. This pricing transition affects any infrastructure software OR hardware company. I picked on Oracle because it has a greater dependence on infrastructure software than any of its peers. IBM can price the software component as part of a larger, services-led solution sell. MSFT revenues and profits are still dominated by the desktop not the server business (though cloud pricing could cause the desktop business model to unravel). And SAP, as an applications company, still has the relative safety net of user-based pricing. That is far less often tied to metering.

      Regarding Oracle metered pricing on AWS, it’s only permitted for Oracle Standard Edition One – which is a severely crippled, low-end version of the mainstream enterprise database. The RDS service specifically requires all of the larger versions to be acquired via a traditional contract.

      Share
  4. mohammedfarooq Saturday, March 10, 2012

    Oracle is striking partnerships with Iaas providers like Savvis, Terramark and others with preferred/discounted licensing models for the cloud. It is still not metered through. They are requiring Oracle Virtual Machine as a condition to get preferred pricing. If you choose to use Oracle on VMWare, the restrictions are very constraining and expensive. I am assuming that they are trying to move their virtualization technology to get control of the metering layer below their database software before transitioning to the metering model in a controlled and phased approach. Something to think about…Nice post !! I have faced many of the Oracle licensing challenges discussed in this article when onboarding enterprise class apps to the Iaas platforms using our cloud brokerage and governance service.

    Share
    1. George Gilbert Sunday, March 11, 2012

      Mohammed –

      You touch on a really interesting point. For years Oracle refused to offer product support for customers on VMware. I think they gave special dispensation to very large customers. With respect to ownership of the virtualization layer, I think you identify a critical insecurity for Oracle. If they don’t control the virtual machine, someone else, likely VMware, knows how many physical processor cores are supporting the database at any given time. That means Oracle can only **indirectly monitor utilization and billing**. With Oracle VM, they have that direct control. However, they still require licensing for peak capacity on Oracle VM, just like on their engineered appliances.

      Share
  5. Campbell Webb Saturday, March 10, 2012

    http://cloud.oracle.com

    While not metered per se, the service offers small/med/large configurations, scale up, monthly billing.

    Share
    1. George Gilbert Sunday, March 11, 2012

      Campbell-

      Oracle’s Public Cloud appears to be the closest they’ve come to offering metered pricing. Although they haven’t actually announced the pricing level yet (*and that is a significant detail*), they did say it will be based on a monthly subscription with elastic capacity. By comparison, Amazon offers the granularity of hourly pricing.

      But the biggest limitation to this offering is that is Oracle-only software, at least for now. If you want to build an enterprise app or *especially* a Web site, you won’t have access to all the richness of software from other vendors and most importantly from the open source community.

      But this does bear watching. Microsoft’s Azure originally only had Microsoft software and now it is growing to encompass open source software as well.

      Share
  6. Jeff Schneider Saturday, March 10, 2012

    Exadata does not feel very ‘cloudy’, however, Oracle on AWS RDS does have the traditional cloud attributes. Although it’s early, cloud.oracle.com also feels like it’s heading down the same cloud-aligned path. Best to include them in an Oracle cloud discussion?

    Share
    1. George Gilbert Sunday, March 11, 2012

      Yes about cloud.oracle.com – Campbell asked a similar question above to which i added a comment.

      Share
  7. Lots of really bad info in that article…

    -$2.8M would be almost 120 intel cores of Oracle Enterprise Edition at list price. There are well known websites run on half of that.
    -You couldn’t run a transactional DB on anything other than Oracle very successfully on 120 cores
    -Unless your hosting provider is bleeding money, its going to cost you a *lot* more over 3-4 years than 2.8M to host that. The electricity alone to run that is far more than what you quoted.
    -Could you come up with a segmented architecture to run in all on mySQL? Sure, you’d also probably spend an extra $20M in development costs to do it and manage it.
    -As far as running a 2 node cluster on RAC – you’d only need standard edition and you’d closer to 30K in licensing then, not 900K.
    -Oracle already announced its offering hosted “amazon” like public cloud databases for EE. You can let cat of the bag on your secret… cloud.oracle.com
    -No serious company on the Web outsources its hosting. Its a great start up model, but the cost benefit is waaaaay past there when you’re at 2.8M in database licenses. This was Ellison’s knock on Force.com, you end up having to take a proprietary format that you *can’t* take off a public cloud (this is also a problem for Azure)

    Share
    1. George Gilbert Sunday, March 11, 2012

      Mr. DBA (I can’t address you as stupid, since you’re comments are well informed):

      “$2.8M would be almost 120 intel cores of Oracle Enterprise Edition at list price. There are well known websites run on half of that”

      I think I know where your processor count comes from. $2,800,000 / $23,700 / core = apx 120 cores. Let me try to explain my math, since the pricing rules can be complicated.

      Amazon’s largest EC2 database instance is “Quadruple Extra Large DB Instance” and it has 8 virtual cores at the equivalent of 1Ghz 2007 class Intel Xeons. So you take that EC2 database instance and then you convert the 8 virtual cores to 4 “Processor Core Equivalents” because of how Oracle accounts for 1Ghz 2007 class Xeon cores.

      Enterprise Edition costs $47,500 per processor core and another $46,000 per processor core combined for the Real Application Clusters and Cache options – so that you can run in a configuration similar to a mirrored mySQL with Memcached. That totals apx $400,000 *per node*. So the two nodes in a cluster together total apx $800,000. That gets upsized by a factor of 2.5 to account for peak capacity. Even if the application runs at that peak 10% of the time, you have to pay Oracle *up front* apx $2,000,000 plus 22% for first year maintenance.

      – Regarding the running cost of this configuration, the baseline configuration is a total of 16 virtual cores or 2 EC2 max database instances. Amazon’s fully loaded charge for running mySQL in that configuration is only $5.20 per hour. So unless Amazon is bleeding red ink, the electricity is included in that. Similarly, if you upsize that to peak capacity, it’s still only $12.00 per hour.

      – Regarding the comment that Standard Edition would suffice, you couldn’t use the Cache option for memory-resident acceleration like Memcached, or any of the other Enterprise Edition functions for that matter.

      – Regarding Force.com, apparently they are hard at work at moving off Oracle as the database for their PaaS offering as well as for the core Salesforce.com application itself. Once upon a time Scott McNealy used to boast that big applications needed big Sun boxes and he frequently pointed to Salesforce.com. Then they moved to commodity x86 boxes.

      – Regarding building the site or application on a sharded mySQL architecture, you’re right. It is more difficult than using a proper RDBMS where you leave all physical management of the data to the DBMS. However, the default stack and application pattern for Web applications that emerged over the last 10 years was LAMP on sharded mySQL databases. Driven by Big Data requirements, It’s now evolving even further to include even more scalable databases labeled (or mislabeled) NoSQL or NewSQL.

      – Regarding running on Oracle Public Cloud, see my response to Campbell Webb above. We still need to see pricing and how non-Oracle software can be accommodated.

      Share
    2. George Gilbert Monday, March 12, 2012

      Forgot to mention one other thing in my reply. According to the pricing documents I read, Standard Edition allows you to add the RAC cluster option but you can only run a single node. You can’t run it that way in a cluster. In other words, the option is there only so that you can have a common programming and admin model with other servers running Enterprise Edition.

      Share
  8. Whilst I agree with the disruptive nature to the addiction of the capital sale of traditional ISV’s and the entire value chain of the software business, you are incorrect in suggesting that Oracle does not offer metered DB or other products in its stack. Checkout http://aws.amazon.com/rds/oracle/ . Whilst not heavily promoted, it’s available and I use it.

    Share
    1. George Gilbert Sunday, March 11, 2012

      Yes – they do offer RDS for Oracle. But see the next comment after yours and my reply. It’s not the full product. It’s severely restricted in functionality compared to the full Enterprise Edition.

      Share
  9. Are you aware that Oracle is available today on Amazon as a metered service? Try doing a web search for “Amazon RDS for Oracle”. This whole line of argument seems to be based on a false premise.

    Share
    1. George Gilbert Sunday, March 11, 2012

      Anon1 – this is what made writing the post so fun. You have to wade through *dozens* of documents on Oracle’s Web site as well as on Amazon to figure out the what they’re trying to obscure. Amazon RDS for Oracle only permits you to license Oracle Standard Edition One using metered pricing. This edition is severely restricted in functionality. Quoting from their collateral:

      “Oracle Database11g Standard Edition One delivers unprecedented ease-of-use, power, and price/performance for workgroup, departmental, and web applications on single servers with a maximum of 2 sockets.”

      This edition cannot be run in a clustered configuration at all – including a write-master and read-replica-slaves. More important, you can’t add *any* of the Enterprise Edition options that make sophisticated applications possible. This starts with Real Application Clusters. It includes options for caching, data compression, data encryption, synchronized standby machines for reporting, partitioning, and a dozen other features that are core to the enterprise edition offering.

      Share
  10. Nice post! Really oracle has secrets.

    Share

Comments have been disabled for this post