Capacity of distributed infrastructure
Beyond the lack of governance, I would like to make the following points:
- Commoditized computers used in distributed applications are cheap, and it might be less expensive to waste 50% of a 1 CPU system, vs. wasting 50% of a mainframe's CPU, hence, less incentive.
- The capacity of systems calculated just by using the CPU utilization might not reflect that the memory or I/O could be at full capacity.
- Maximizing the utilization of systems used by a single distributed application is more complex because of the extreme "NUMA" nature of these architectures.
I agree however that by unifying the architecture, and using a central resource management tool coupled with dynamic provisioning tools is a must, and should improve the agility of the datacenter, allowing to implement "just in time provisioning".
I would suggest also that, in order to improve the aggregated utility of a disitributed infrastructure, it is necessary to run more than one application on any given server, by using some virtualization techniques, for example.