Infrastructure is rarely linearly scalable. To be truly linear, there should be means to increase the capacity of the infrastructure in a straight line. There always seems to be a sunk cost, that is, there always seems to be a step function in the graph.
Take blade servers as an example. In concept, blade servers should provide linear scalability. One buys the blade chassis and then adds blades to the chassis as computing requirements increase. However, a blade chassis has a finite number of slots. Once the computing requirement goes beyond that number, one has to buy a new chassis. Assuming 100 slots, blade number 2 to 100 has the same price, but blade 101 is really expensive.
Although blade servers appear to offer the promise of linear scalability, in practice they are no different than normal big servers. Because of the sunk cost of the chassis, customers always buy a chassis with all the blade slots already filled.
What if you could actually achieve real linear scalability. To do so, the backplane would need to be able to grow linearly. Or better yet, there would be no backplane at all.
What if you could just add server after server? No additional chassis to buy after you reach the 101st blade. No additional switches or routers to buy. How would that change how we plan and manage our infrastructure?