Okay, I wanted to talk about IEEE's recent event
honoring SPARC; but, before I even get started, I should address a couple of points about the title.
First: when I’ve written before about our technology history, I've sometimes been asked, “Should you really be talking about how old
such-and-such a thing is?”
Not even a problem here. As far as current enterprise processors are concerned, SPARC
’s still the new kid on the block:x86
Mainframes? Still around, as you know. Built on an architecture
announced in 1964
And yes: I see that hand waving frantically in the back: “WAIT! WAIT! But why does the title say ’1989
’? Wasn’t SPARC introduced in 1987
Well, yes. Yes, it was. Gold star for you!
But it’s been not dead
A PROUD TRADITION OF BEING NOT AT ALL DEAD CONTINUES
In 1989, I'd just joined Sun, after several years as a customer. I was of course up on all the news about the brand new SPARCstation 1
, and I was excited to learn what was coming next.
Which was a full-frontal assault from our competitors.
Not to blame them; I mean, here we were, the fairly-new guys (only 5 years old at the time SPARC was announced)—and we dared to come out with an entirely new processor architecture?
And this is where, two years in, we started to hear people (well, our competitors) say: SPARC is dead! Never going anywhere.
HP was probably the leader in death-knell pronouncements, because they’d just bought Apollo, which had a line of "SPARC-killer" workstations based on the PRISM
processor. Those faded out right about the same time I found out where the coffee supplies were in the office.
But that’s OK, because HP had a new line of “Snake
” workstations, which would totally crush SPARC! Those did a bit better; the underlying chip hung on long enough to become the processor that HP killed in favor of Itanium
. SPARC-crushing: not so much.
Next up was DEC Alpha
, which was—you guessed it—going to be the end of SPARC. (My favorite DEC presentation tidbit from back in the day: the transition from VAX
to Alpha was going to be E-Z ...because they could both read the same plaintext data files ! I’m not making this up.) Alpha’s still around, technically, as is Itanium. SPARC killing, though: not achieved.
But there’s more to this story than just sheer survival—SPARC has a habit of becoming the leader in technologies other companies had to scramble to match: RISC
computing, extreme chip multithreading
, and now Software in Silicon
. SPARC stays not-dead the way any technology does: continuous innovation.
OK… but what’s the next reason for it to be not-dead?
How about four?
FOUR REASONS TO BET ON SPARC
1. Oracle and this other guy named Larry
A lot of people were skeptical—very skeptical—about what was going to happen to Sun’s hardware business after the acquisition. (That’s OK; some of us at Sun were kind of interested in that, too.)
Larry Ellison answered that pretty boldly on the day the deal closed, laying out a roadmap for SPARC/Solaris systems. How’d we do on that? Not too bad:
Oracle’s strategy is to be in a business for the long haul, and nothing shows that more dramatically than what has happened with SPARC since then—moving beyond “not dead” to “holy…”. Oracle’s intensified efforts on both SPARC and Solaris are demonstrably paying off, in a big way, at the same time our main enterprise competitors seem to have lost their way.
—Reduced Instruction Set Computing—was a then-obscure concept that Bill Joy championed in the mid-‘80s, resulting in Sun developing the SPARC architecture. The idea’s very simple: bet your silicon real estate on the instructions that do you the most good. As big an idea as that was then, it’s even bigger now.
Here’s why: it’s the best way to deal in silicon with a reality we’ve been living for decades: most compute tasks today are ludicrously over-provisioned on a single processor. An architecture based on simplified processing units lets you put more flexible compute on a single chip, for less money, less power, less cooling.
By putting more compute units on a single chip, instead of deploying a lot of low-performance systems in your data center, with redundant support hardware, you can now put all that compute in a more cost-effective system. You drive down cost yet further, and you drive down latency, the real enemy of performance today.
3. Writing to a chip? You still do that?
Back when I joined Sun, rolling out a new processor architecture was considered to be audacious. Building a whole new ecosystem was challenging, and optimizations for a particular chip/OS combination were hard-fought battles only won by the most brilliant developers.
This hasn’t completely gone away, but it’s not what you, the developer/operator, do any more. You’re going to chunk away at much higher layers of abstraction, and let our campus full of folks in Santa Clara
take care of the optimizations. Our deal with you is: we’ll build and co-engineer awesome hardware, operating systems, software and developer suites
, and you use it to build, run, and maintain the services you need.
Bottom line: if you can do it, you can do it on SPARC. And what we do makes it better.
You didn’t think I was going to miss that one, did you? This is a Solaris blog, after all.
The combination of chip multithreading and virtualization has turned the concept of data center performance exactly upside down. Instead of worrying about how to get enough power to a single application, the challenge is to how to get enough applications onto the power you have.
SPARC/Solaris solves this challenge brilliantly, with an unmatched combination of enterprise dependability and cloud agility. The strengths that have made Solaris the “go to” OS for enterprise computing continue to get even better, and now we bake in things like OpenStack
, built-in virtualization
, and compliance tools
to make your life easier, and your applications rock-solid and wicked fast.
ONWARD TO THE SECOND INNING
There's a tendency to think that we're always at the end of the game, when we're really only in the first inning. If you look at the history of SPARC, you see competitors who wanted customers to think the game was over, before most of the crowd had even settled in their seats.
A few days ago, IEEE celebrated the beginning of SPARC, and that gives us a chance to talk about where that's going to take us. Keep in mind, this isn’t the 9th inning of the compute era. SPARC’s more than just “not dead”—it’s just getting rolling.