Another Oracle Open World keynote, another pack of lies from Larry Ellison. Bad enough that the still blind (dangit, why does my prescription have to suck so much?!) guy is actually writing a blog post about it. Before people start spreading more lies as truth. Mind, I’m not there, so this is an incomplete dismantling of the falsehoods. But there’s some real ugly ones.
And yeah, being a Sun guy for as long as I have been, well. I feel obligated to share the real facts of things, so folks aren’t reliant on spin, revisionism and unicorns.
Oracle now <3′s Fujitsu, honest!
This one is about as bald-faced as it gets. Larry doesn’t like rebranding deals, because he’s stuck with what he can get from somebody else. Larry likes to manage all the technology, to get his hands dirty – he’ll tell you that himself. He’s a technology guy.
Larry doesn’t like the Fujitsu deal. But he also can’t escape the Fujitsu deal. And it is absolutely not a new thing. In fact, the Fujitsu relationship is a very ancient tradition. See, Fujitsu is the other big half of the SPARC alliance. Oh, you thought it was a product? It’s not. It’s an acronym – Scalable Processor Architecture – and the trademark is actually owned by SPARC Internaional, Inc. The two primary members since 1986 when the SPARC V7 reference architecture was released? Sun and Fujitsu. UltraSPARC Architecture 2005? Sun and Fujitsu again.
(Of note: POWER ISA is also a similar arrangement, with the majors being IBM and Motorola, now Freescale.)
Add to this the absolute fact that Solaris customers have been 100% dependent on Fujitsu for non-T series SPARC systems (which still make up the bulk of Solaris installations worldwide) since January 2009, when Sun EOL’d the last of their UltraSPARC IV+ systems. While Jonathan “I’M NOT A TOTAL IDIOT!” Schwartz sent Sun on the failboat journey that was Rock, they leveraged a tiff with Fujitsu over support contracts to gain exclusive sale rights to Fujitsu’s SPARC64 hardware in North America under the Sun brand.
Every single M-series system is exclusively designed and built by Fujitsu. And customers running M-series chose them for a reason; they are not running applications and workloads that can be handled by the T3/T4 family’s meager performance, nor do they translate to the Exa* unicorns. And if Ellison axes the deal with Fujitsu, he immediately loses a very, very lucrative revenue stream as every single M-series customer abandons ship. The annual software maintenance on even an M5000 can easily exceed $150,000 – and that’s almost pure profit when the system only gets replaced every 3-4 years.
Oracle software running on Fujitsu hardware is new and a big deal!
Could you tell a bigger whopper? Probably not.
In fact, Fujitsu is and remains the only other company in the world which has a license to sell SPARC systems with Solaris. 9 out of 10 Solaris systems in Japan are built part and parcel by Fujitsu with an OS from Sun^WOracle. And Oracle inherited that deal, which isn’t believed to expire until 2013 at the earliest.
But here’s the thing: Sun software has been running on Fujitsu metal for over 20 years. Not “about,” not “close to,” but over. The first Fujitsu CPU to ship running SunOS? The 1986 Sun 4100 system board used in the 4/260 and 4/280, which was equipped with a Fujitsu SF9010 IU coupled to a Weitek 1164/1165 floating point unit and ran SunOS 3.2. The first actual Sun designed processor wasn’t introduced until 1992, the STP1020 AKA SuperSPARC I. (Sun has never actually fabricated their own silicon – they have always been a fabless design shop. Most CPUs were actually manufactured by Texas Instruments and TSMC, but were designed and validated by Sun.)
So it should come as absolutely no surprise to anyone at all that Oracle’s software products have been running on Fujitsu metal for decades. Three decades, actually – they added support for Unix environments in Version 3, released in 1983, which included Fujitsu metal. But the relationship goes much further than just “running on Fuji metal.”
Set your wayback machine to 1999: Oracle 8i. The i was for “Internet” before it was for “obnoxious Apple marketing gimmick.” Back then, Sun had yet to ‘open source’ Java, so you know they were close as close can be when Oracle included their native JVM with 8i. (You can probably dredge up some of the old ‘The Network is the Computer’ co-marketing materials with Sun/Oracle, too.) In parallel, Fujitsu was shipping their PRIMEpower family with the new SPARC64 GP processor, which unlike the SPARC64 II, had a simple enough MMU to run vanilla Solaris. And was one of the fastest SPARC V9′s of the day. No surprise, they ran a lot of Oracle 8i installs too.
Honestly, putting things back into the silicon on SPARC isn’t an Oracle initiative or new either. It’s Sun’s fault it got taken out in the first place! The SPARC64 II had an incredibly complex and effective memory management unit that handled most memory management tasks, even at the OS level. That was too complex for Sun, which forced HAL Software Systems to develop the SunOS derivative SPARC64/OS specific to the SPARC64 I and II. Fujitsu had to take out the advanced silicon functions in later SPARC64 family CPUs so they could run Solaris!
Oracle’s First Cloud^W^WThe First To The Cloud!
This one just made me cringe. Seriously? Ellison actually tried to pull this one off?
Look. I could go on for six weeks about Sun’s cloud attempts, dating back to the McNealy era. “The Network is the Computer” campaigns, on-premise ‘hosting’ efforts, managed hosting, Sun Compute Grid (2004ish as I recall), Sun Cloud (the 2007 reintroduction), and those are just thinking of the ones that made it out into the wild where customers could look at them.
Oracle claiming to be the first anything cloud at this point is easily the most insane and delusional thing possible. As much as I have slammed Sun over the past 5+ years, you will never catch me denying their brilliant moments over the past 20 I’ve worked with them. And Sun was shipping – not marketing, but actually shipping – the great granddaddy of cloud, Sun Grid Engine in 2000. Not a typo – 12 years ago. Oh, and they offered a free version for Solaris and Linux. And open sourced it in 2001.
12 years ago. Is it the cloud you know and [love,loathe] today? Not any more than you or I are our paleolithic ancestors. Evolution sometimes leads to relationships that are difficult to recognize, but SGE is the predecessor to all, like it or not. (While there’s predecessors to SGE, they’re all dead. SGE still yet lives. Sorta.)
That evolved into Sun Cloud, which like it or not, your modern cloud also derives from. Users uploaded their data into the Sun Cloud as a bundle which contained their data and paid for how much CPU time they used. They could also select an application from Sun’s Application Catalog (introduced in 2007) which let ISVs and developers publish their applications directly, and other users simply click to select a prepared image which included the appropriate binaries and libraries.
Larry Ellison took the axe to Sun Cloud in 2010 and yanked the plug with virtually no warning on March 7, 2010, trying to throw the whole affair down the memory hole.
“Exadata X3 1/8th rack is $200K” and “way faster.”
I shouldn’t even need to cover this one, but let’s be honest here, I need to because not everyone is an Oracle customer. And this is the biggest whopper of them all.
First of all, unless you’re a member of the Oracle Religion, you know better than to believe a word of Oracle’s benchmarks. They’ve been slapped with more fines and bans for false and misleading advertising – specifically writ benchmarks – in the past year than IBM, HP and Dell combined in the last decade. When Larry says it’s faster, what he means is “we cherry picked a workload we could do well at, then we gave our system at least 20 times as much RAM as the competitor and probably more than twice as many cores.”
That doesn’t mean Exadata isn’t faster – just that stating it as a fact is lying. Period.
That $200K will get you nothing more than a bunch of x86 boxes. Not even particularly impressive x86 boxes for the price. You can get a better deal on HP Gen8, IBM xSeries, Dell PowerEdge, and Cisco UCS than that. And if you’re buying Exadata X3-2, let’s be honest, you’re not buying it for the hardware. You’re buying it for the software pitch.
Well, based on Oracle’s current pricing sheet, to actually do anything with that Exadata X3 is going to cost you $200K (or less) up front. That’s a fair and truthful statement, to say the hardware is $200K or less. (Ignoring that they’re risking another investigation for selling to customers below GSA list.)
Based on various calculations around the web from competitor analysts, DBAs, and customers, Exadata X2 cost you a cool $1.1M for a rack of metal – and another $5-7 million dollars per year in licensing and support to use it. As Larry used to say constantly, Oracle is a software company not a hardware company. And they charge like it. And as Pythian pointed out with their excellent analysis of Exadata X2, the more storage and cores you put in it, the more the licenses you need.
Know why customers didn’t applaud ‘same price for twice the hardware’? That’s exactly the reason why – Larry could give away Exadata (but he won’t, because Oracle gives nothing for free, ever) and make it all up on the licensing. Increasing the specifications might increase performance, but it has a much bigger impact on driving Oracle’s rent-seeking^Wlicensing revenue.
I don’t even want to touch the revisionist history portion of the presentations. Suffice to say, this is barely scratching the surface. And if you do your own digging, you’ll quickly find out that Unca Larry’s unbreakable wall of bricks is just cheap paint on an empty refrigerator box.