Re: EXTERNAL: OT: Linux vendor survey results

From: Matthew Zito <>
Date: Mon, 9 Dec 2013 13:39:36 -0500
Message-ID: <>

Note: looooong response.

You can ascribe a variety of motives to Oracle when it comes to OEL. Dave, I think your educated opinion could generally be ascribed to positive motives - Oracle reacting to a perceived anticompetitive threat from RHEL.

I think the broader perspective to put OEL in is to look back to Oracle circa the early 2000s. The economy is tanking post dot-com bubble, Microsoft is getting into the database business wholesale, the world is moving away from giant big-iron boxes from Sun and IBM and to commodity Intel boxes running Linux and Windows, and Oracle is under price pressure from its customers to justify their support renewals.

So Oracle takes a step back and realizes that in reality, they're just getting a small piece of the pie of one of their database server sales. Money is going to EMC for storage, Sun for hardware and OS, Veritas for storage management and clustering, Quest for performance management and admin, Legato/IBM/Veritas for backup, and so on.

Oracle then embarked on a systematic campaign to own the entire stack. I know that a lot of these capabilities existed prior to this time period, but Oracle invested a ton of money in beefing up RAC, introducing ASM, ADDM, Enterprise Manager, RMAN, and so on.

I ran the numbers once back in the day, I wish I still had the spreadsheet around, but whereas oracle used to get something like 40-50% of the total cost of an enterprise database server, they now get upwards of 80% - that's an estimate, I don't remember exactly what it works out to.

The last piece they didn't control was the OS. Again, thinking back to circa 2003--ish, Red Hat wasn't the overwhelmingly dominant player it is today. I remember walking through a major NYC bank in 2003 and their talking about their giant RHEL pilot and how skeptical they were about a commercial enterprise Linux distribution. If I remember correctly that pilot was considered a failure and it was another year or two before they took another look.

Oracle, at this time, saw Linux primarily as a low-cost alternative to Windows. I remember conversations I had with folks from Oracle at a management level who dismissed Linux as primarily a lightweight OS that they could use as a foil vs. Microsoft.

So why not back Linux? If they want to push commodity hardware, there's only two OSes that are going to run on it - Linux or Windows, and Windows opens up SQL Server and a tightly integrated stack. Linux it is.

Fast forward a few years, and Red Hat becomes the dominant player, at least in the US. Red Hat now is in a bit of a tough position. The original Red Hat versions tracked the generic kernel releases very closely. One of the reasons they released an "enterprise" Linux was that their customers wanted a more stable kernel that changed less often. In this case, their "customers" are really three different groups:

  • their end customers - big banks, who have to certify new OS builds on all of their hardware, software stacks, internal processes, management tools, and the like - and need to be able to get support for older builds for a long period of time
  • Hardware vendors - IBM, HP, Dell aren't going to rewrite their drivers every six months to keep up with the rapidly moving stock Linux kernel. If Red Hat and SuSE want to be supported on a broad range of hardware, they need to play ball with the hardware vendors
  • ISVs - companies like Oracle and everyone else were also pushing Red Hat to standardize on a set of kernel interfaces.

If you haven't paid close attention to Linux kernel internals over the last 10 years, they change *a lot*. Red Hat, SuSE, IBM, and others all came to a consensus that stock kernel behavior can change quite a bit internally as long as it doesn't break any of the core POSIX notions or binary compatibility. However, memory management, kernel tunables, driver architectures, scalability enhancements, that's all fair game. This allows the community to move at the speed it wants in the spirit of a free and open source platform.

Those internal changes are exactly the sorts of things that scare the hell out of Red Hat's three constituent parties - after all, changing memory allocation rules arbitrarily in between patch releases would hardly be good for Oracle. Consequently, the decision was made that the RHEL kernel is going to be based on a standard kernel, with patches added to target specific functionality with consultation from RHEL's three customer bases.

Over time, this gets tricky - as someone who has built the RHEL 3 kernel from scratch back in the day, it was the 2.4.21 stock kernel with something like 150 patches on top of it. But Red Hat was willing to take on that complexity in order to provide their customers with the consistency they required.

But has Red Hat "forked" the Linux kernel? By some definition, yes. It was done, though, at the behest of their customers and partners, who needed the stability that came with having a consistent kernel tree across the life of a major release of RHEL. It simply doesn't work any other way.

Now, back to the business angle here. As Red Hat grows in dominance at an OS level, they themselves start to round out their capabilities. Instead of just offering an OS and quietly collecting their annual subscriptions, they start moving up the stack. They acquire JBoss, an open source J2EE application server, and Netscape directory server. They also start heavily investing people resources into other open source projects. They launch a cluster server product, and a global filesystem project that actually works. We're up to around 2006-ish here.

Meanwhile, Oracle themselves heavily moves out of their core database business into applications, acquiring everyone and their mother in the J2EE application space, weblogic, JD edwards, Peoplesoft, and so on. Again, their strategy is to have the complete stack - the second vertically integrated technology stack in the enterprise (the first being msft, obviously).

Except now there's a third player in the game. Red Hat has the OS, it has some of the supporting administration tools, it has a java app server, it has an excellent reputation amongst users. There are plenty of open source software companies for them to acquire all the way up to enterprise apps themselves. Even if Oracle's products are better than the open source alternatives, it forces price pressure on them.

Red Hat is nowhere near being a competitor to Oracle in 2006 - but heck, if JBoss takes off that hurts Weblogic, Red Hat acquires EnterpriseDB, suddenly there's a third horse in the race.

So why not cut them off at the ankles? Red Hat's whole business model is predicated on being open source, why not simply download all of the RHEL source packages, recompile, add whatever value-add they want, and sell support at half the price of RHEL? Even if they break even on the support, it hurts Red Hat and protects their vertically integrated stack.

In fact, I found an old article:


Red Hat’s own acquisition last week of JBoss – another open source company whose products compete with Oracle and IBM – could shake up alliances in the Linux world and provide another reason to act, said Mr Ellison.

“Now that Red Hat . . . competes with us in middleware, we have to re-look at the relationship – so does IBM,” he said.

On Red Hat’s growing influence and its ambitions to reach beyond the Linux operating system, he added: “I don’t think Oracle and IBM want another Microsoft in Red Hat.”"

So Oracle has every incentive to prevent Red Hat from getting any further traction in the market.

Yes, I know that with the UEK, Oracle has actually added enhancements over the stock RHEL kernel and good for them. I know that most/all of them have been given back to the community.

However, some of the accompanying capabilities of the UEK, like the ksplice online kernel patching functionality is not available to RHEL users - just OEL users.

All of this behavior and history, in totality, would seem to indicate that Oracle Linux is less about believing that Red Hat has done a disservice to the OSS community, or Oracle's belief in Linux as an open source ideal, and a lot more about:
- offering a vertically integrated stack from top to bottom (see: exadata) - preventing a low-cost, open source competitor from infringing on their market
- "because we can shank Red Hat from a position of safety"

I don't ascribe malice to any of those things, just good ol' fashioned corporate strategy. But it sure ain't warm and fuzzies.

Thanks for the patience for anyone who bothered to read this whole thing.


NB: I might have some dates screwed up on some of this, it all happened over a long period of time. Hopefully the gist of the progression is clear regardless

On Mon, Dec 9, 2013 at 11:18 AM, David Roberts <> wrote:

> I definitely don't want to expand this thread into a broader Oracle and
> Open source discussion, where Oracle's sins are apparently legion.
> But in general Oracle doesn't understand/play well with the Open source
> community.
> I don't believe that is the case with Linux.
> The appeal of Linux to Oracle is obvious.
> The cost of a solution will be the cost of hardware + software stack.
> Where competitors like Microsoft are pushing their 'expensive' operating
> system as a mandatory component of any SQL server installation, having a
> 'free' operating system available under an Oracle database stack, then
> Oracle can get a bigger slice of any same priced pie.
> The view from Oracle is that Red Had didn't play fair.
> Red Hat have effectively (in Oracles view at least) forked the Linux
> Kernel, this has allowed Red Hat to overcharge for it's support by
> providing a barrier to entering the market.
> Red Hat enhanced their kernel and keep it as a separate development tree.
> When new mainstream Linux kernels are released, Red Hat evaluates the
> changes and then if appropriate migrate them into their kernel fork.
> When Oracle encountered problems implementing their database on RHL that
> required patching of the Linux Kernel, in Oracles option Red Hat didn't add
> these patches into their kernel in a timely manner, typically Red hat would
> wait for the patches to be included into the mainstream kernel and then
> port them back into the Red Hat kernel.
> It was for that reason that Oracle adopted the mainstream kernel,
> enhancing it and offering it as an alternative the the RHL kernel.
> I suspect that it would be possible for oracle to produce proprietary
> drivers for Linux, although I don't really see the motivation for that.
> I suspect that incorporating the new driver to the Red Hat kernel might
> not actually be straight forward.
> I can see the motivation for Oracle to not produce a RHL version of the
> driver, but I suspect that Oracle will contribute the source to the
> mainstream kernel and Red had can choose to adopt it from there.
> If Oracle does retain this code as proprietary, then I can see this
> damaging the relationship with the Linux community which I believe was
> relatively good up till now.
> Dave
> NB. I would classify all the above as educated opinion and definitely not
> fact!
> x

Received on Mon Dec 09 2013 - 19:39:36 CET

Original text of this message