For years Sun has been harping that "the network is the computer", which sounded pretty good as a marketing slogan back in the 90s when the industry had barely recovered from "SAA" and "DNA" (ten bonus points for anyone old enough to remember 1) what those stood for and 2) whose architectures they were!). While it was more marketing slogan than architecture, and the others were slightly more architecture than marketing slogan, in the end they all occupied about the same locus in the universe of ideas - a somewhat amorphous term onto which folks could hang all sorts of actual realities.
Now Sun has graced us with "Redshift Computing" (originally courtesy of CTO Greg Papadopoulos). Outside of the unfortunate need to be real careful when typing this relatively new computing term, I actually really like it. It sounds all scientific and kind of big idea-y, and in a certain sense it is.
Sun is making the basic case that scale will dominate everything, that a very small number of providers (five to be exact, well sort of ...) will become the dominant providers of that dominant scale, and that Sun will dominate the dominant providers of that dominant scale. (no wonder they coined a simpler term - not to sure "dominant cubed" would have done too well)
At any rate, it's a seductive argument at first glance, except like so many other architectures / marketing slogans there is a real strong tendency to gloss over some enormous realities ("there's no need to set a plate for that enormous elephant standing in the middle of our dining room honey ... after all, I don't think he'll like what you made for dinner").
Let's start with one simple awkward reality - of the examples (Google, Salesforce.com, Amazon EC2 & S3, Ebay, & Twitter) that Sun VP Peder Ulander uses in an interview with David Berlind (check it out as part of a very good post (When there’s only five giant computers left, will one or more of them be Sun’s?) the largest (Google, amazon EC2 & S3) are using commodity (i.e. not Sun) gear, one (Ebay) has a very high margin business so infrastructure TCO is not all that important at this time (of course it may be at some point, but that's a post for another day), another (salesforce.com) has used Sun gear but had some notable outages, and another (Twitter, hosted by Joyent) is arguably somewhat cost-sensitive. And this is Ulander's own list! For those keeping score that's two decent examples out of all of them, and my guess is that both of those (salesforce.com and Twitter) would have been better off with an actual commodity implementation. In fact, fellow Appistry blogger MarkSU showed just how to use a fabric to do a commodity Twitter). Furthermore, as Isabel Wang notes here Twitter has been down often, had trouble scaling, and well really needs to find another way to deploy! Even the best example, salesforce.com, has had its stability and scale problems (check out this post for more info).
But I digress. The point is that the best examples Sun uses to justify the contention are, on the whole, not using Sun gear! And of those that do one has an enormously profitable revenue model that eases the focus on infrastructure costs, one has had some very notable problems, and the other doesn't really work very well.
Oops.
Still, I do like the notions that the need to scale is upon us, that our future scaling needs are enormous in ways that we have trouble comprehending right now, that we may well be looking at some very different ways to be deploying our apps in order to handle that scale, and sure, all this scaling is going to have some very interesting impacts on the economy.
Even if Sun may not have this future all sewn up like they're contending, and even if this future might contain just a wee-number more than five computers, and if maybe, just maybe there might be a few major issues to resolve to make sure that you're ready to scale ad infinitum, at least they have some things right. Besides, I think Papadopoulos makes some very interesting, albeit incomplete points. And I do like the name.
So take the time to read David's post, check out his interview with Ulander, and check back tomorrow for my next installment on Redshift Computing.
[note that I have corrected this post to reflect the usage of Sun equipment by salesforce.com, as pointed out by Andreas Schwarz in the comments - thanks Andreas!]
Technorati Tags: commoditization, datacenter, redshift computing, scale data, scaling Web 2.0
Hi Bob, I handle public
Hi Bob,
I handle public relations for Greg, and saw your post on Redshift. Speaking for myself (and not Greg), I agree that Sun isn't there yet. Then again, nobody else is, either, or Google would buy hardware from them.
What we're doing is directing Sun's $2 billion / year in R&D to solve these issues -- what we call "brutal efficiency at massive scale." For example, check out the efficiencies in Project Blackbox (www.sun.com/blackbox). That's just a starting point.
Regardless, definitely agree with your main point, and looking forward to Part II tomorrow!
Andreas
P.S. For the record, Salesforce.com: http://www.sun.com/smi/Press/sunflash/2001-12/sunflash.20011204.1.xml
And Twitter:
http://twitter.com/blog/2007/06/under-hood-at-twitter.html
So while there's certainly room to improve, our record is better than suggested above ;)
Hi Andreas, Good points all
Hi Andreas,
Good points all - thanks for your thoughts regarding my initial post on redshift computing. Correction noted with regards to salesforce.com - must be due to my advancing age ... heck, I was happy to remember dinner last night! Having said that, it did bring to mind this post by Dan Farber which followed a couple of notable salesforce.com outages.
At any rate, thanks for joining the conversation, & pass the word to any who have something to contribute. Thx!
Bob
Hi Bob, Thanks for the
Hi Bob,
Thanks for the mention in your Blog following the story. While it may seem awkward that I mentioned a number of companies that are not using Sun technology, I did so deliberately. Understand that it would be easy to make a 100% Sun argument, but then it is only looking at the world through Sun-colored glasses and the reality is that this is a trend that is happening with or without Sun. Interestingly enough, when you think of the companies that have exploited high-bandwitch plays and truly bet on the network - the ones that have used commodity-type, whitebox solutions have all run into serious issues because the systems weren't built for scale, space, power, etc. And while the Google, YouTube and MySpaces of the past could manage through that then, the competition is so hot now why risk it? The reality is that the edge of a 5 computer world is more "in your face" type of a statement and we will see a number of pools and farms across the globe. And of those new companies sprouting up, I see Sun having a solid position as the partner of choice given our drive to invest in technology growth and platforms at all layers of the network.
Regardless, thanks for the post. Keep it going with Appistry, cool technology enabling this new world of computing.
Cheers,
peder
Sun seems to be suffering
Sun seems to be suffering from split-brain syndrome at the moment. The creative side of the brain realizes that a shift is underway and that scale is one vector of that shift. The logical side of the brain, however, seems unable to divorce itself from the concept of Sun as a "server" vendor.
As evidence of this, take a look at the Black Box project. The creative side of the brain came up with a very interesting idea - a mini data center delivered anywhere in the world as a standard shipping container. Great. Unfortunately, though, the logical side then screwed it up. Insisting that servers are the critical deliverable, more than a third of the space in the Black Box is dedicated to human access to the servers. You can bet Google didn't make such a mistake with their container project.
At Google scale servers are like light bulbs in a warehouse. When one fails, it's a non-event. It's simply powered down automatically. No one scrambles to reboot it or look for a replacement. And, like the light bulb, it'll get addressed as part of a batch by the janitorial staff some weeks or months later.
As I understand it, redshift
As I understand it, redshift computing seems to contains several related but separate ideas. There's the notion that hyperscale computing (exemplified by Google having zillions of servers) requires new IT practices. Then there's the notion of hypergrowth, infrastructure growth far in excess of Moore's Law replacement levels. Okay, that's fine. But so what? Presumably Sun wants to target this massive growth opportunity, but it seems like a fancy way of saying "we want to sell where demand for our products is highest." When was that not the case?
Also: will Sun learn to love selling lots of commodity-priced hardware? They seemed to have great reluctance facing the Linux threat, which was very ugly to their traditional business model. Do they plan to out-Linux Linux in the realm of super-cheap commodity servers?
Peder, Bert, Jason - Thx for
Peder, Bert, Jason - Thx for the discussion and perspective. I'll have the next post up later today. Hope everyone had great weekends!
- Bob
[...] the last five systems,
[...] the last five systems, Sun is not off to a good start. Although Sun spokesperson Andreas Schwartz pointed out how salesforce.com announced in 2001 that it would be running its platform on Sun’s Starfire [...]
Post new comment