In powering up the Sun Grid, Sun Microsystems is not aspiring to operate computing grids, but rather to demonstrate what can be done.
Sun needs to foster the idea of grid computing if it hopes to sell its servers, storage, and middleware (including its Grid Engine grid software). It is for this reason that the company has lashed together a few thousand computers running its Solaris operating system, Grid Engine, and its Java Enterprise System middleware to create a prototype utility computing platform that it can show off and test its subscription-based pricing models on.
After talking about the grid at length, the company has now actually turned on the electricity in the Sun Grid and is taking its first customers. While Sun wants to make money directly and immediately with its Sun Grid computing and storage facilities, there is more to the project. Only by demonstrating the facilities to companies and stakeholders that consume computing power can Sun drive the use of (rather than the theoretical idea of) utility computing.
In the same way that the electricity generation companies of a more than a century ago had to demonstrate that centralized, utility power generation is better and cheaper than generating electricity for their own consumption, Sun and other players who want to supply utility computing solutions have to show why this idea is practical now as well as inevitable somewhere between now and the far, sci-fi future. This is why Sun has built its grid and is charging $1 per CPU per hour for computing and $1 per gigabyte per month for storage on it.
Sun’s CEO and chairman, Scott McNealy, said at the Network Computing 05Q1 launch event in Santa Clara that the company is being practical as it rolls out portions of the Sun Grid in co-location facilities – Sun is not a data center operator nor has it any desire to be one. The centers that make up the Sun Grid, which is comprised of its Sparc and Opteron servers, are located nearby Sun’s own factories and distribution centers.
The company has approached the situation in this way as it is experimenting with the idea of doing its server burn in on the Sun Grid, and making its servers do useful work while components are being tested for infant mortality. This is smart, and it echoes what PC maker Gateway tried before it shut down its retail stores a few years ago. Gateway partnered with grid software innovator United Devices to make a vast amount of computing resource available for something useful.
In trying to put the whole utility computing concept in perspective, Mr McNealy showed his usual wit. Our industry is as screwed up as any industry in the world, except possibly healthcare, where they lose every patient – eventually, he quipped, so the bar is pretty low. Mr McNealy and Mr Schwartz, who will be making presentations to Wall Street analysts this week, are well versed in the ideas Sun has been espousing and that they need traction.
Now is the year we have to show some results, and we know that, he said. We need to deliver consistent growth and consistent profitability. He added that the utility computing model Sun is fomenting is not new – that water, sewer, transportation, electricity, and communications systems had all gone from a burst of innovation, through hype, to a crash, and then standardization that eventually pushed them into utility models.
Mr McNealy believes that, because Sun owns most of the components of a utility grid, something he says only IBM can also claim, it can be the low-cost supplier of computing and storage capacity for its own grid and, more importantly, the ones that it wants its partners to build.
He says with the pricing that Sun is using, when you roll in the costs to administer in-house systems and factor in the low usage (about 10 to 15% of the capacity) that most companies have, Sun is offering capacity at anywhere from one to three orders of magnitudes cheaper than what IT shops pay to build such capacity themselves.
While neither Mr Schwartz nor Mr McNealy would talk about how many CPU-hours or GB-months it would need to sell to make it break even, the fact that Sun Grid will be used to burn in systems that are actually sold to customers means the cost to Sun should be quite low. It is quite profitable, said Mr McNealy – something that should get Wall Street’s attention.
There are two main problems with grid computing. One is that it is really only suitable, for now at least, for number-crunching jobs typically run on supercomputers or parallel clusters. Immediate applications can be found for the Sun Grid running simulations in the financial, oil and gas, and biotech industries, as well as for digital rendering and design automation.
For a decade, Sun has run a cluster of distributed Sparc systems running Solaris and its EDA software to create its chips. This cluster has 13,000 processors and a 97% average utilization. So when Mr Schwartz and Mr McNealy are a little bit smug about how hard grid really is and how competitors such as Dell will not have an easy time, maybe they have a point.
Electricity is cheap, but it isn’t easy to generate, warned Mr Schwartz. Dell is to Sun what a piston ring is to a cab ride, Mr McNealy responded when asked about what Sun would do when Dell would, seemingly inevitably, try to move in on Sun’s grid turf.
The second, well-known barrier to the adoption of grids is cultural. IT departments, divisions, whole companies, and governments have bought and managed their own computers for four decades. It will be hard to get them to let go and move to a utility model.
But the force of history is on Sun’s side, not the IT managers’. There used to be a job called chief electrification officer. We don’t need that guy any more, we just plug into the wall. But there is still a need for electric generating systems and electricians.
