The Register has an article on this. As I noted in the previous post (the linked article really), ASP reborn is a hard model to make work. You have to deliver what customers want, and do so inexpensively.
Sun Microsystems has killed it’s once high-profile utility computing experiment, Network.com, which let customers buy computing power by the hour.
The company revealed it’s no longer accepting new customers after four years, saying parts of the business and technology model “were not in the sweet spot”. The 13 customers and 48 applications using Network.com are will be offered continued service.
Amazon is succeeding at this in spades. Sun originally had the right idea … then someone decided to make Network.com into a massive Solaris marketing tool rather than a resource you could run what you needed.
We were working with Sun then, and tried to get some of the local people to comprehend that this was in fact a bad idea. We pointed out that this was a barrier to entry for existing code, for commercial code. We tried to interest some of our customers in this platform, and they were, as long as we could run it as a linux cluster. Which we couldn’t.
See the problem?
Well, this wasn’t the only problem. The other one is cost.
Take a server, lets call it a $5000 server, with 4 cores (back when they built this), and try to set a price for its usage. Sun set $1/CPU per hour. Sounds reasonable, right?
Lets use a simple (and incorrect) linear depreciation model for the hardware. So that for the 3 years (26250 hours) of depreciation, this $5000 is resulting in a loss in value of about $0.19/hour. So in order to recover the price of the hardware, you need to charge at least this amount, assuming a constant usage model.
But wait, you note. That server has 4 CPUs. So on a per core basis, we are looking at something like $0.05/hour. This doesn’t mean that Sun was making 2000% margins. There are fixed and variable costs, the system’s utilization might be 20% (magnifying the needed per core price by 1/utilization). Lets assume for the moment that they had 20% utilization. This makes the per core costs about $0.24 per hour. Estimating the fixed/variable costs as approximately doubling these costs, the actual approximate costs to Sun were likely about $0.48/core per hour.
So $1/hour isn’t terrible. Its actually not unreasonable.
(yeah, we have worked through this before for VCs and others trying to build something like this here in Michigan several years ago … before Network.com)
But the problem is that reasonable machines now are $2500/8 cores, so the costs have gone down by effectively 4x. In this light, the business model for them had been untenable by pre-building the infrastructure. Indeed, you can dollar cost average your infrastructure costs by buying and standing up on demand.
What Network.com was fighting was the internal IT department buying their own servers and standing them up. The costs for that are in the noise, and the depreciation model while inexact, works well enough to estimate the cost per hour.
In which case you start fighting the cost-benefit analysis battle. And this is one I don’t think Sun could win for this sort of scenario, against the internal groups.
Now add a much lower cost provider basically making a product that enabled you to stand up what you want, and not turn their product into a marketing vehicle for their other lines of business. This is what Amazon did.
Suddenly Sun had 2 major competitors here, one with an internal bias and preference to keep control local, and an external competitor offering a general platform.
This is one of those sink or swim momemnts. Not to be confused with a moment to fall back on ones corporate ego. Using it as a marketing platform for Solaris was a mistake.
Their plan to run Linux as a guest OS is at least a step in the right direction. But this model requires you pull as much cost out of the hardware as possible, and reduce the cost/complexity of the software stack. Having the customer own/run their own stack does that, and inexpensive machines to run them on handles the other issues. Sun would do well to study EC2.
Viewed 3939 times by 592 viewers