DISQUS

Pete Grillo · 3 months ago

There are lots of attractive benefits to SaaS/cloud computing. However the advantages may not be what they first appear to be. The costs, as you point out, are anything but ‘free and getting cheaper’. No denying the cost of storage and bandwidth has continued to drop. Simple math says spreading the fixed costs – startup, development, etc – across a higher number of users equates to lower costs and greater profits.

I wish it were this simple. First, things change as user counts grow. Back-end databases that purr like a kitten act very differently when user counts go up by a 100x increase. Or 10,000x. Look at Twitter, a great service that is well capitalized and still has uptime issues. My point is that data centers don’t run themselves. Development and tuning go on forever. So where do you put employee costs with a hosted system? If you place (hide?) them in development they appear as one-time costs. It could be argued that the staff running a hosted service belongs in COGS. The numbers look very different when you start adding employee costs to the raw storage and bandwidth costs. Still attractive, but not ‘free and getting freer’.

In my experience at iterasi the benefits to SaaS are primarily convenience and version control. It is convenient for a customer to be able to access their account from any computer anywhere in the world. It is convenient for the SaaS provider to deploy products, new features and bug fixes in one place without hoping that all customers get the email or the CD and actually install the patch. Having sold software in a variety of configurations, this one alone makes SaaS the platform architecture of choice.

And, btw, we built our own data center in a co-location facility. Why? Because despite what conventional wisdom tells us, we could not run the number in such a way as to make cloud computing come close to the cost of rolling our own.
pete

Thorsten Claus · 3 months ago

Hi Pete, Thanks for your comment, and I fully agree! And S3 is of course not really up to par in terms of latency, so no actual real-time data, please :)

One of the common slides that I present says that scale of (real-time and no-so real-time) information logistics is a design problem, no a resource problem.