The Database Is the Computer

So many smart technologies have been created for and embodied in the successive generations of IBM‘s midrange platforms over the past four decades that it is hard to single out any particular idea of merit. If you put a gun to my head and made me choose, I would probably ask for the ability to single out one hardware idea and one software idea, which is a kind of cheating, I admit. But what else would you expect?

On the hardware side, I think that single-level storage, which masks the complexities of where data is in cache and main memories and on disk drives within the system was a brilliant insight. This made programming so much easier, and made it possible for IBM to use a mix of different hardware technologies in an asymmetrical way that ultimately yielded excellent price/performance using modestly powered processors–in this case, a variant of the Motorola 68K processor (which I have always believed that the CISC processors inside of the AS/400 were) and a bunch of 68K chips that wrapped around them, yet which presented a single address and storage space to applications and their related files, as if it were simply one giant chip with one giant memory space. This is a brilliant insight, and if there is any surprise to me it is that all computing is not done this way. (It just goes to show you how might makes right most of the time when right ought to make right. (Yeah, I know, I kinda added two hardware ideas there. But again, this is me you are talking about . . . .)

On the software side, the truly distinguishing insight that IBM had was to not only bundle a relational database with the operating system, but to essentially kill two birds with one stone and make the relational database the actual file system on the computer itself. This idea was way ahead of its time, unfortunately, in some ways. But the System/38 and its CPF operating system, which was perfected as the AS/400 and its OS/400 operating system, had the right idea decades ago. And it is funny to me how Microsoft was attempting to create essentially the same effect with its “Longhorn” server implementation of Windows, which will be called Windows Server 2008 when it ships early next year and which was supposed to have an integrated database-like file system called WinFS. WinFS was gutted out of the operating system because it was too difficult for Microsoft to deliver.

I find it less amusing that in the mid-1990s, as PC servers were becoming more prevalent, IBM had to abandon this database as file system approach in OS/400 because the overhead of storing ASCII files inside of DB2/400–the unofficial name of that integrated database that has never been given a proper moniker–was making the AS/400 a poor performer when it came to client/server computing. So IBM gutted OS/400 and propped up the database on a rejiggered implementation of the OS/2 High Performance File System, which we know today as the Integrated File System. With this move, OS/400 and now i5/OS look exactly like every other operating system, which distinguishes between files stored in one file system and data stored inside the relational database. Over the long haul, I think computing architectures will move back to where IBM was with the System/38 in 1978–the database is the computer.

Database software maker Oracle certainly believes this, and attending the Oracle 11g database launch last week in New York made it abundantly clear to me that Oracle understands the brilliance of what IBM accomplished with the design of the System/38 (just as Oracle was being founded in Silicon Valley in 1977) in terms of technology and with the AS/400 in terms of improved technology and market acceptance.

Just to make it clear how Oracle has come to dominate the relational database market, dig this. Oracle’s databases now have a bigger footprint in the IT space than DB2/400 has, and that is a very big accomplishment. One that no other database has ever done before, in fact. Charles Phillips, Oracle’s president, said at the 11g launch last week that Oracle now has 275,000 customers. According to my best guess, the peak size of the AS/400 installed base of customers (not machines, but unique companies) was around 280,000 in 2000, when Y2K and some very big deals in the late 1990s pushed the installed base to that height. IBM could have as many as 215,000 unique customers today, but only says that the installed base is north of 200,000 customers officially. Any way you look at it, the base of customers using DB2/400 as their database has shrank by somewhere from 65,000 to 80,000 customers, and, without question, Oracle has been the biggest beneficiary of those lost OS/400 and i5/OS footprints. I can’t tell you how many times I have heard of a merger where a company using Oracle on Unix has been acquired by or has acquired a company using OS/400 and DB2/400 and the IBM i5/OS platform loses the competition to survive. I don’t know what the Oracle 6, 7, and 8 installed base was in 2000, but I am fairly confident that it was nowhere near 280,000 unique customers.

As Oracle has matured as a company, it has done what all operating system companies have done–keep integrating successive layers of technology into the software it controls. In Oracle’s case these days, and particularly with the new Oracle 11g, which starts shipping in August on Linux boxes and later in 2007 on other platforms, an operating system that supports Oracle looks less like an operating system and more like a kernel that supports the Oracle database operating system. (You can read all about the nifty stuff inside Oracle 11g in this article that ran in our sister publication, The Unix Guardian, last week.) There are many very smart technologies in this new database, and that sustained innovation explains why Oracle has nearly a 50 percent revenue share in the relational database market. With Oracle 9i RAC, database clustering was built in at the database level, and this was improved with Oracle 10g and has been inched toward perfection again with Oracle 11g.

Moreover, Oracle understands the importance of controlling the software stack as much as possible, and if it cannot control the underlying operating systems on which the Oracle database runs, the company has done a marvelous job with integrating Java virtual machines, creating application servers, content management systems, parallel file systems, integrated database clustering that is transparent to applications, and development tools that ride on top of the database or alongside of it. And, of course, Oracle has built up its already substantial application software business by acquiring PeopleSoft, J.D. Edwards, Siebel Systems, and myriad smaller firms. Finally, with Oracle’s Unbreakable Linux clone of the Red Hat variant of Linux, Oracle can now sell a complete stack of Oracle software, from the operating system up to the ERP application suite. Now, Oracle can today do what IBM Rochester and IBM Atlanta used to be able to do for midrange customers: provide a complete, integrated application stack to customers, up through applications. As IBM has been unplugging itself from customers in the past three decades, leaving it to customers and relying on services, Oracle has been amassing pieces of the stack. Guess who is more profitable?

Let’s take a look, and as I write this sentence, I actually don’t know. IBM had $91.4 billion in sales in fiscal 2006 (ended December 2006) and brought $9.5 billion to the bottom line. That works out to profits of 10.4 percent of revenue. Wall Street, which assesses the relative strength of a company through its market capitalization (the aggregate value of all of its common stock based on its ability to bring in profits), has IBM worth $162.3 billion, or about 1.8 times annual sales for 2006, at its current stock price. In fiscal 2007 ended in May 2007, Oracle had sales for the year of just a hair under $18 billion and a net income of $4.3 billion. Oracle brings 24 percent of sales to the bottom line–2.5 times the rate of Big Blue. And that, more than anything else, explains why Oracle has a market capitalization of $104.8 billion as we go to press, which works out to a factor of 5.8 times sales in fiscal 2007.

Some people never learn. Larry Ellison, one of Oracle’s founders and the man who was smart enough to sell the first relational database, based on ideas taken right out of IBM’s research papers, to the Central Intelligence Agency back in the late 1970s to get Oracle rolling, is not one of them.

Next week, I want to revisit the idea of creating a grid variant of the System i platform, an idea I floated two years ago and that bears some re-examination given the success that Oracle has had with its Real Application Clusters. My original story is in the RELATED STORIES section below if you want to recall what I said back then. The advent of user-based pricing on the i5 515 and i5 525 machines as well as the creation of the i5/OS Application Server this spring make for some interesting possibilities for the System i servers in a grid computing environment.

RELATED STORIES

Oracle Says 11g Database Is Better, Cheaper, and Faster

Forget Oracle 10g. Let’s Talk About i5/OS V5g

                     Post this story to del.icio.us
               Post this story to Digg
    Post this story to Slashdot