Original URL: http://www.theregister.co.uk/2007/12/05/sun_redshift/

Sun warns of voracious, collapsing data centers

McDonalds may sell them

By Ashlee Vance

Posted in Servers, 5th December 2007 23:46 GMT

Sun Microsystems loves to think ten years ahead of the market. Such technology philosophizing is natural for a research and development heavy. As of late, however, Sun has started to claim that its grand vision of computing's future will start to take hold not in the next decade but rather next year.

During 2008, Sun expects to see the full-on arrival of what it calls RedShift applications. These are the types of software jobs that require more gains in processing power than Moore's Law can provide.

At the moment, the high performance computing (HPC) set come to mind as the most obvious RedShifters. These folks turn to massive clusters and all manner of accelerators to obtain huge speedups in application performance. In most cases, the HPC customers must place an emphasis on writing parallelized software that can spread across many systems well and/or tweak their software for the specialized accelerators to realize the performance gains they desire.

The software work done by the customers and vendors gets around the end of the GHz boosting era. Coders can no longer count on Intel, AMD and others to crank up the clocks on their chips but must instead find a way to take advantage of the extra cores being added to each new generation of processors.

Another class of RedShifters comes from the software-as-a-service set. You can think of companies such as Amazon, eBay and Google/YouTube in this group.

More than Moore

Looking at Google, for example, and you find a company that ignores Moore's Law as we've come to know it. Google uses rather slow chips to power its core services. The ad broker, however, makes up for underwhelming individual horsepower per server by linking thousands of servers together and writing code that can tap into these cores well.

According to Sun CTO Greg Papadopoulos, 2008 will mark the first time that RedShift type companies start to consume the majority of available compute cycles.

This is a very forward thinking position to take. It means that more attention and money will go toward building new, massive scale services rather than contracting BlueShift services such as, say, payroll.

To support the RedShift era, companies will start building data centers that defy belief, according to Sun.

We're talking about 50,000 sq. ft, 5,000kW centers now and 500,000 sq. ft, 50,000kW centers over the next ten years.

But, again, Sun is talking about the now rather than the next ten years and predicts that at least one organization - perhaps a government lab - will begin work on a 500,000 sq. ft data center in 2008.

In addition, Sun VP Subodh Bapat warned that 2008 will bring a data center failure of unprecedented scale, causing not only tremendous pains for users but also possible "national security issues." So, there's that to look forward to, as our data centers reach enormous proportions.

The rather comical bit about Sun's predictions is that the company spends so much time thinking about these things but fails to demonstrate any business around the ideas to the public.

The RedShift-style data centers receiving the most attention these days happen to come from Google and Amazon.

Google avoids the major hardware suppliers altogether by building its own servers and networking gear. Meanwhile, Amazon shot past Sun, HP and IBM by creating a quite popular utility computing services which allows companies to run their applications and store their data on Amazon's systems.

Sun has tried and mostly failed to create a similar utility program for its customers.

In many ways, Sun miscalculated with its effort. The company went after very large customers in the financial services, oil and gas and media fields, offering them CPUs and storage on an hourly rate. As it turned out, Sun struggled to navigate red tape associated with security and compliance issues with these shared services and failed to get much traction with its grid.

Amazon, meanwhile, targeted much smaller organizations and has done well with its approach.

Is it sad that a retailer beat a visionary server vendor to the punch here? Yes.

In addition, there are an awful lot of $600m data centers going up these days. Google, Microsoft, Yahoo and others are building them. And yet we see no concrete details or even hear rumors of Sun's involvement with the mega data centers.

During the last build out boom, Sun found itself at the center of the whole ordeal. It owned the telecommunications companies' accounts and made tons of money as a result.

We see no evidence that Sun is doing something similar this time around.

Does this mean it's doom and gloom for the forward-thinking Sun?

Well, no.

At long last, Sun has revamped its entire server product line and looks to have some compelling future gear on the way, coming from the server, storage and software sides of the house. We tend to think that a grand reset button has been pushed at Sun, and the company now has a clean slate to work with as it goes after the RedShift business.

Sun's talk matches the realities we're seeing in the server market more closely than any other major vendor, and it has products to support the vision. Now the big question remains as to whether or not Sun can get talk and action working at the same time.

If not, who knows, we might see McDonalds beat Sun to the next major shift in computing. ®

Register editor Ashlee Vance has just pumped out a new book that's a guide to Silicon Valley. The book starts with the electronics pioneers present in the Bay Area in the early 20th century and marches up to today's heavies. Want to know where Gordon Moore eats Chinese food, how unions affected the rise of microprocessors or how Fairchild Semiconductor got its start? This is the book for you - available at Amazon US here or in the UK here.