Ellison Answers Financial, Technical Questions
Ellison Answers Financial, Technical Questions
SAN DIEGOWednesday at Oracle AppsWorld here, Larry Ellison was one busy CEO. In between meeting with customers, he managed to fit in an hour-long keynote, a briefing with various groups of the pressincluding a trade press briefing attended by eWEEK reporter Renee Fergusonand another long Q&A session with two dozen financial analystsalong with eWEEK.com and a few other general reporters.
The session with analysts covered a wide range of topics, from Oracle Corp.s worst mistakes during the go-go 90s to the future of computing, Oracles view on Linux, and how the company plans to compete with Microsoft Corp. and other enterprise software vendors.
Ellison, along with Oracles other executives, made a point during the gathering of financial analysts to promote the companys adoption of "Lintel" (Linux and Intel) servers as a low-cost way to deploy large databases. Heres what he had to say about the true "entrance into the information age," brought about by low-cost clusters of servers, organized into a grid.
"Theres been a 40-year quest to build the biggest computer in the world. At Amdahl [where Ellison started his career] we built a faster mainframe than IBM.
"Most applications are designed to run on only one machine. How many IBM DB2 database machines can run SAP? One. Whats the maximum number of SQL Server machines? One. Whats the maximum number of Oracle machines? 128."
Ellison went on to describe the performance and cost differences between the current model of enterprise database computingwhere you load a database on an expensive SMP boxand Oracles vision of grid computing, an array of cheap Intel servers running Linux and Oracle 10g in parallel.
"We found that the sweet spot [for these grid systems] is dual-processor machines, not four. Six-thousand-dollar dual-processor Lintel machines. The $2,500 processor [3.2GHz Pentium] is twice as fast as the $50,000 processor [in large SMP servers].
"Its 40 times the cost/performance of big iron, and the grid of small machines is much faster than the big system. Its real on-demand computing. Sam Palmisano at IBM is having trouble explaining on-demand computing to IBM employees. Id say thats because they dont have on-demand. The technology for them to deliver on-demand computing is a purchase contract. The technology we use for on-demandplug in another machineis software. You need more capacity, you add another cheap machine. Its a fundamental change in the computing business.
"We support companies that make very fast, low-cost, small machines. We think Lintel is very attractive. I talk with [Sun Microsystems Inc. CEO] Scott McNealy all the time, hes a good friend, and Sun is on board with this. I think they will be the first to deliver very fast AMD [Advanced Micro Devices Inc.] machines running Linux. They will miss the 32-bit Linux show, but they want to be first with 64 bits. I think they will be the first to deliver 64-bit Linux.
"We think this has implications not just for the database business, but for the entire industry."
As for Linux, Ellison doesnt see Red Hat Inc. turning into the next Microsoft and taking over the industry, but he does see the open-source operating system beating Windows in the data center.
"I think its impossible for there to be another Microsoft. Lets say Red Hat decided to increase the cost of Linux by a factor of 10. IBM and Oracle would have their own distributions of Linux the next day. Wed cut them off. They just cant do that.
"Or wed quickly switch customers to SuSE. The Linux intellectual property doesnt belong to Red Hat. Further, Red Hat doesnt service many customers. Red Hat relies on us to service Linux for them. Why do our [enterprise] customers buy Linux? Well, they arent willing to rely on Red Hat to fix the system if it goes down. They are willing to rely on us. We take full responsibility for those systems. I dont think anyones going to trust Red Hat for that.
So how many Linux distributions do we need?
"I think one Linux version controlled by one company wouldnt be good. One hundred wouldnt be good either. Whats the optimal number? Twos the absolute minimum, but three or four isnt bad. We fully support that.
"We think a lot of people misunderstand Linux. Everyone focuses on open source and free. Well, theres some open-source and free stuff thats not very good.
When the topic turned to Linux versus Windows, Ellison was outspoken in defense of Linux in the enterprise.
"Linux is a much better data center operating system than Windows. The Windows design point is for having one Windows machine in every hamburger store and every bank branch. Its a distributed model. Linux works better when you have 500 machines in one data center. Its a better product than Windows, like Apache is a better product than IIS. We saw Apache wipe Microsoft out of that market entirely, and I think Linux is going to win the server wars. There are even people at Microsoft who think that it is inevitable."
Making the transition to
The conversation moved over to Oracles new 10g database. It took two years for 50 percent of Oracles customers to move to 9i, how will the company accelerate the transition to 10g?
"Our ability to automatically manage the disk farm in 10g is so good that a lot of DBAs [database administrators] will want to get there as soon as possible. Were letting them say, Here are 50 spindles, you Oracle figure out how to spread the data. Its a manual process in DB2 and SQL Server. Its tedious and fraught with danger as you move data around. That feature alone will cause people to migrate to 10g.
"Its also much faster. Weve seen something like a 50 percent performance increase on our applications. Big [performance] increases on the query optimizer too. Were the only database that can usefully employ multiple machines on real-world applications. The only ones with greater clusters that work. Were the fastest on one machine. We work on a grid. We have lower cost of ownership.
"But that feature to automatically manage your disk pool is so compelling. Dont believe me, all the DBAs say it is so compelling and it just works.
Ellison repeatedly emphasized Oracles lower cost of ownership.
"This is an enduring theme. Youll see us continue lowering the cost of ownership. Get rid of third-party software and get rid of labor. Wherever you have labor, you have errors. Automating the labor out is absolutely crucial, because you get paid twice. And thats the best way to compete with Microsoft going forward, a much better cost of ownership."
Oracle has built a number of components into 10g that are also provided by enterprise software companies like Veritas, Cognos and other vendors. Ellison spent a significant part of the Q&A talking about why they now offer those tools, and the companys strategic planwhich seemed to borrow heavily from Microsoft.
"Oracle 10g operates better without Veritas than with Veritas. The backup part aside. Using the Veritas file system and volume manager is an unnecessary cost when using 10g. Its more expensive, more complicated and we dont recommend it. Our goal is to provide the most reliable and simplest system at the lowest possible cost.
"By getting rid of the [Veritas] volume manager it lowers the cost to our customers. I think we have a much better volume manager. Its simpler and free. We think the same thing is true of our management system. Having them tightly coupled makes it more secure and more reliable. And its less expensive from a software cost and a labor cost. We are trying to eliminate the requirement for expensive software add-ons. Were doing the same with business intelligence. That lowers the cost, and makes the system faster and more reliable. On our application servers, we built business intelligence into the database to eliminate the need for Cognos or Business Objects. We have to continue to lower the cost of our system, by eliminating the price of costly add-ons.
"We have to maintain our advantage of performance over Microsoft, and reliability over Microsoft, but also lower the cost of ownership and make our system more secure. We dont require Business Objects. We dont require Veritas. Well be adding our own grid control too, making our system more reliable, more secure and at a lower cost.
"Whats Oracles ambition? To do some of the things that Microsoft has done, to sell products at a lower price against the competition. Veritas charges a lot of money for volume management. Our product is free. When you assess the size of the market, you realize we are selling at a much lower price point than the competition. We are trying to sell a higher volume of product at a lower price. We compete with Tibco, Cognos, Business Objects, Web Methods and Veritas. We sell at a much lower price, not unlike what Microsoft has done with PC software."
But what about the new Real Application Cluster (RAC) service, which manages the array of servers and storage devices in 10g? Why is Oracle charging for that?
"Short answer, because we thought wed make more money charging for it than not.
"We charge for options based on performance and reliability, we dont charge for features. We decided to charge for RAC because it is so new and highly differentiated. Maybe at some point we wont charge for it, but for now its such a unique feature that it wont slow adoption.
"Weve got so many customers that have bought all the Oracle theyll ever need, and we need some database technology to sell them that they dont already own. Its so different, so unique, no one else has it, so we thought we could make some money on it. Id rather lower the price of the database and still charge for RAC."
Ellison continued to discuss the challenges in bringing 10g to market, and how RAC evolved.
"One of the things about grid computing thats important is not the database software, but the management software. Weve had to come out with a whole suite of management software to manage the grid. When you go from one mainframe to 64 machines and you have a bug, how do you patch 64 machines? How do you start up 64 machines? We had to build a whole set of management software called Grid Control to help deal with a large number of small machines, as opposed to a small number of large machines.
"Weve been working on RAC for a very long time, and now we have much better management software. In the old RAC days, a typical RAC would be two to four machines; now its 64 or more. Its that much more important to have the right management tools. Are we developing management tools to go after the companies in the tools industry? Yes. But we had to. Are we going after Veritas? Yes, but we had no choice. When were doing innovative things, sometimes you have to go beyond what we think of as a database, both up and down the linea complete system from high-level management tools to the volume manager itself."
Ellison spent a long time in his keynote describing the concept of a "customer hub," where all customer data resides in one place. He used the example of a new worldwide credit verification applicationwhich tracks every person in the world capable of receiving creditas just one of the customer hubs Oracle is helping to build. But what about hubs beyond the customer?
"The customer data hub is just a bit misleading. Its really a customer/product data hub that answers the question, Who are my customers and what do they own? But theres personnel data in there too. You have to know all your salespeople, all your customers and all your products. You have to include territories, your entire sales force, who is supporting those customers and whos the credit person. When we say global customer database, its a huge percentage of your business database.
"The next one is supplier, and then employee, and I cant think of what else there is."
Learning from past mistakes
Ellison went on to detail two major Oracle failings during the 90s: creating a single sales force with one rep per company, and building a huge internal consulting and services organization that competed with the channel.
"One of the single worst mistakes Oracle ever made was thinking that the customer wants one person selling everything. I thought that was strange, because the customer doesnt have one person buying everything. I think the one-person sales force model was a tremendous mistake. We separated applications from the database sales force, which we should have done a long time ago.
"We are an infrastructure company. Eighty percent is infrastructure, and 20 percent is applications. But were unlike any other application company in the world. Were the only application company with a huge infrastructure business thats separate. Are customers forcing us into integration? People are talking a lot about doing different types of integration with message passing. That style of integrationpassing data from one application to anotherhas been around for 25 years.
"Were talking about a different way to integrate, around a common data model and database, instead of around a message backbone. Large customers have always had a need to integrate, but Ive always thought that standard message passing technology was enormously expensive to use. I think IBM Global Services would agree with that. Gerstner said that for every dollar in product revenue, they brought in five dollars for services. We think that ratio is way out of line.
"Where IBM makes their money is in system integration. If one of the companies that begins with GeneralMotors, Electricdecides to build a customer data hub [with Oracle], the likelihood of choosing our consultants to do that is very small. General Electric would probably pick Tata Consultancy Servicesone of the biggest Indian consulting and outsourcing companiesbecause they watch every penny.
"One of the most interesting things that happened last year, which I learned after the fact: GE Medical implemented our eBusiness suite without me knowing about it. Well who did this? Tata. An awful lot of this work will be done by third-party consultants. I dont see us ever having, nor do we want, the scale of consulting business to do these big projects. Our primary business: We are a software technology company.
"Seen over the past few years, Oracles been flat to down. But its our consulting and education business thats down. Software and licensing is a lucrative business. We dont want to compete with system integrators like we did in the 90s. Thats another big mistake we made. The big system integrators were recommending SAP and Seibel because they saw us as a competitor. We dont want to be a competitor."
The recent management shake-up at Oracle, where Safra Catz and Chuck Phillips were promoted to president, and Jeff Henley to chairman, was a topic of conversation around the show. Heres Larrys response to why he did it.
"It was all designed for one simple focus: to have our senior executives spend more time with customers. Thats priority one, two, three, four and five. Jeffs been our financial officer for a while, and weve had conversations about the best ways for him to spend his time. And thats with customers. The more time Jeff spends with customers, the better input and the better management decisions we make. They [Katz, Phillips and Henley] are going to be spending much more of their time with customers, so we have better information from our customers on what were going to emphasize and what were doing next."
Ellison went on to discuss why he gave up the chairman role, but stopped himself as he began to discuss what, exactly, his role is to be.
"Ive been chairman of Oracle for half the time Ive been there. [Ellison reels off a list of other Oracle chairmen from the past.] Im not an expert in accounting, Jeff is. Im not up on all the new rules. Jeff is. He knows Oracle from the inside. And I cant think of anyone better equipped. But that was Priority 1a. Spending more time with the customers was Priority One."
"And we certainly want to keep Jeff around too." (laughter).
There was a lot of talk about pricing, particularly since the cost of 10g hasnt been announced yet. Ellison was critical of industry-standard pricing models based on processors or users.
"I think all pricing models have flaws. Theres always an edge condition. The best model is an employee pricing model thats annual. Lets count employees, and then give a company all the Oracle you can eat forpick a numbersay $100 an employee per year.
"Its very difficult for our customers to even count processors, or users of systems. Ask a large company how many processors they have! They just dont know. Its a much more convenient pricing model to count the number of employees you have each year. I think thats a much more reasonable approach to pricing, and thats a direction wed like to go. Right now you buy the product, and then pay a 20 percent subscription service each year. My favorite situation would be to have no new license sales and all subscription services."
Ellison used his bully-pulpit as an opportunity to chastise the financial analysts for failing to see what he considers a fundamental piece of Oracles business.
"Whats not well-understood about Oracle is that if nothing changes, if we sell $3 billion worth of new softwaresame as last yearour profits go up $600 million. You would say Oracle didnt grow. But because the subscription service goes up 20 percent [of that $3 billion number] we add $600 million in profit. We cant avoid it. We can figure out some way to spend it, I suppose. But the most profitable part of our business is our subscriptions, not new software sales.
"That subscription is a very interesting line. They think its part of our service business. Its not. Its part of our software business. And there is zero cost. Zero service. If software sales do increase, its more!
"We think our margins can expand very substantially. I dont think we want to control the expansion [what some people call investing]. I think we have more sales capacity. Well have more efficiency, and more use of telesales. Theres plenty of existing capacity to grow our sales. So you should see a top line growth of sales. Youll see substantial margin expansion."
Larry went on to suggest that a 50 percent margin is feasible, although Oracle has yet to hit 40 percent. After he left the room, newly minted Chairman Henley took the stage and tried to clarify Ellisons discussion of marginfocusing on 40 percent first. Henley was then joined by new presidents Katz and Phillips, who clarified the pricing discussionsaying that a completely subscription-based pricing model is a long way off.