Originally appeared in MSI Magazine
January 1, 1999

Bypassing Authority

As we’ve noted previously, history often runs in circles in the computer industry. When we noted that the NC was simply the latest incarnation of the X Terminal (see MSI, January 1997), we predicted that it would likely suffer a similar fate. Fast forward a couple years, and NCs didn’t exactly take the market by storm. Instead, we’ve been pacified with sub-$1000 computers and Wintel promises that Zero Administration for Windows and Wired for Management would each take huge chunks out of total costs of ownership. That is, if we could wait for Windows 2000 or beyond.

Now it’s the database’s turn to repeat history. Oracle and Sun—charter members of the ABM (Anyone But Microsoft) coalition, have announced a new OS-less version of Oracle 8. Specifically, Sun will lend Oracle a small bit of OS kernel around which Oracle could wrap and optimize its database. So, when you buy that Sun Enterprise Server or Dell PowerEdge Xeon box, you won’t have to pony up for Solaris, or heavens forbid, NT. Instead, you’ll get a database that will perform so efficiently because it won’t have any extraneous operating system calls to distract it.

Forward into the past. Back in the late sixties, Cincom pioneered the revolutionary concept of packaged software. The idea seemed almost laughable during an era where software was considered to be reels of tape.

If we took Oracle’s and Sun’s notion to its logical extent, Cincom’s daring notion actually represented a step back in computer evolution. Because, once you began developing separate, recognizable software, conflict would eventually emerge: What’s best for software execution may not necessarily be best for operating system or hardware design. Performance tradeoffs would grow inevitable.

And that’s exactly what happened. Application software vendors in the Microsoft world are often left guessing about whether their features might conflict with some undocumented quirk of Windows, while UNIX providers are finding themselves deciding the lesser of the evils when optimizing or configuring for HP, IBM, or Sun.

So the idea of a database machine sounds rather attractive. Just serve the data and focus on optimizing I/O, darn it. But, if that was such a great idea, why did CCA’s Model 204 proprietary database box become history? And why is NCR furiously porting Teradata from its native hardware to UNIX and NT?

The obvious answer is the domination of open systems. The OS might not be the most efficient way to run a database, but at least it provides the neutral ground where databases, applications, and networks can all meet.

Even if dedicated database servers made sense, what about applications? Should SAP and i2 also own their own platforms? Fortunately, ISVs have more important things to do. So we won't have to worry about linking dedicated, “open” application platforms through unproven DCOM or CORBA object request brokers and interfaces.

Not surprisingly, the one place where the database server is being touted is for web applications. Although feeding on enterprise data, the web machine generally operates fairly autonomously of enterprise systems. Yet, wasn’t that supposed to be the place where Linux, the new simple, robust UNIX “open source” alternative to NT, was supposed to get its footing?

The question boils down to what problem we are trying to solve. Are we looking for another way to bypass Microsoft, or simply attempting to solve yet another performance bottleneck?

If viewed from a Microsoft replacement perspective, the database server has one important advantage: Unlike the NC, which required sacrifice of assets that were highly visible to end users, nobody will notice any difference if the database server “lacks” an OS. And IT administrators can rid themselves of complex UNIX or buggy NT code.

However, while databases have traditionally been the choke points, as applications grow more distributed and complex, they could become responsible for yet a new class of tie ups. Significantly, enterprise systems management vendors such as CA and BMC have invested large sums for new predictive modeling tools aimed at solving exactly this riddle.

Therefore, before embracing the notion of eliminating an OS layer from the database for your new distributed, multi-tier application, be sure to ask where the real bottleneck will lie. The answer might surprise you.