Wim Coekaerts

Wim Coekaerts' official title at Oracle is principal member of technical staff, Corporate Architecture - Development. But after Larry Ellison learned of his design abilities and Linux skills, he invited Coekaerts to design an Oracle Linux-based internet appliance. Two weeks later, Ellison had his box and a new company (Thinknic.com), and Wim had a new title: Oracle's "Mr. Linux." Oracle Magazine talked with Coekaerts briefly about Linux, Oracle, and the future of development under operating systems.

You've been involved with Linux since the beginning. How do you feel about the progress of the operating system up to now?

It's been very satisfying. When you consider how much effort and manpower have gone into other operating systems and then you see how far Linux has come in such a comparatively short time, it's great - especially since Linux is free and open source and now supports enterprise-level performance.

Why not Linux sooner? Have there been technical impediments for developers like Oracle or database customers?

It hasn't really been that long, but for a database like Oracle's, the OS kernel has to be able to support the characteristic performant profile of the database - and that means clustering technologies. A lot of Oracle developers have been working really hard with Red Hat to address this need, and we're totally ready. The Linux Kernel 2.4 distribution supports Oracle Real Application Clusters, insuring that an Oracle database performs as optimally as it does under other established operating systems. For DBAs, this success is marked by cheaper installations, much lower hardware and maintenance costs, and simplification of tasks.

Weren't Linux clustering technologies addressed and resolved in Project Beowulf?

That's a different clustering model, intended for HPC [(high-performance computing] where complex problems are solved with processing power distributed across multiple nodes - it's effectively parallel processing. Our clustering - what Oracle is known for - is all about data sharing and disk sharing, and that's a completely different problem. So Project Beowulf is a clustering technology built on Linux, but it is not a database-specific, data-sharing model.

What contribution have you and Oracle made to the Linux kernel?

In terms of drivers, there was a lot of debugging and new stuff we looked at.

We looked at asynchronous I/O - the most dramatic improvements in the I/O subsystem are asynchronous I/O - that is, nonblocking I/O in the kernel, elimination of multiple copies to memory buffers while writing to disk, and reducing contention for kernel locks.

We now provide an async I/O library and test suite for Oracle that uses the AI/O code new in Red Hat AS2.1; we also worked with Red Hat engineers to debug the kernel AI/O code. We worked on O_DIRECT patches to allow for 512-byte direct I/O, which Oracle requires. . . . We're also working on watchdog patches that are in the latest Linux kernel.

We made Firewire driver changes to allow for shared disk devices. . . . We also fixed bugs in bus resets on Firewire, and we came up with a Firewire driver to do remote kernel debugging. We allow someone to attach a debugger over the Firewire bus.

But perhaps the biggest project we started last year was to build a cluster file system for Linux. We had already developed one for Windows NT, which is going to be in production sometime in the next couple months. Originally we took that design and started writing the Linux version of it, however Windows and Linux are architecturally so different that the Linux cluster file system ended up being a rewrite rather than a port. The Linux cluster file system will probably be released later this summer.

What this does is make management of a cluster database much simpler because the file system is easier to work with than with raw disks.

What else? Oh yes, collaboration with Intel and Red Hat on the Infiniband implementation. And in keeping with the open-source approach, all of this is, of course, open source.

What advantages do you see for administrators in moving over to Linux?

Three things. First, the cost: It's a lot cheaper. Since Linux is free, the cost of implementation drops significantly because there is no OS tax - no Microsoft tax, as many people call it. Second, aside from the OS being free, Linux allows for lower-cost, nonproprietary hardware solutions, meaning you can deploy with the most cost-effective commodity solution available for you. And with any new low-cost nonproprietary solution, the solution gets cheaper with time. With our Firewire implementation, for example, multiple nodes can share hard disks over Firewire - with the 2.4 Linux kernel. And last but not least, it's open source. Dependence on a single OS "vendor" goes away.

Do you think developers, network administrators, and database administrators are beginning to see the Linux light?

For sure. In fact, within Oracle, there is a huge migration to Linux underway. With the new enterprise-ready Linux distribution, Oracle is moving hundreds of servers to Linux and there will be hundreds more behind that.

So you see a big industrywide migration coming?

Definitely. For DBAs, an elegant bug-free implementation for clusters has been a hurdle, and that has been removed. Perhaps the next-biggest hurdle up to now has been technical support. Every large and successful OS has a vendor behind it who can stand up and say, "Yeah, we will make those changes or we can guide you through this," or, unfortunately, sometimes they say, "That won't be on the menu for some time."

Since Oracle was instrumental in the development and debugging of particular segments of this distribution, Oracle has agreed, with Red Hat, to assume the Linux technical support role for Red Hat/Oracle customers. We want to insure that Oracle products on Linux provide the same performance and stability characteristics as provided on other OS platforms, and it's in our best interest to see that happen. We're anxious to see Linux succeed as the premier Oracle database platform.

What's your take on the UnitedLinux project?

It's a good thing. It will fortify Linux as an industry standard and help to launch it as the operating system of choice. I think consolidating smaller versions will really help the Linux community - and Oracle - and it will make support easier.

Thanks for your time, and congratulations on your Linux success.

Steve Lipson (sblipson@attbi.com) is based in Oakland, California and is a contributing writer to Oracle Magazine.