Pooled resources have been the foundation of virtualization, but in a cloud environment that enables self-service capabilities (another NIST characteristic of cloud computing) you need to design your infrastructure so that your consumers don't adversely affect the other hosted tenants.
As discussed yesterday, the cloud provider (infrastructure IT staff) partitions the Provider vDCs resource allocations with Organization vDCs. These are then presented to the Organizations.
A layer below in vCenter, this sub-divides the resource pools into the organizational structure. The Silver Resource Pool contains 400,000 MHz of CPU and 2,250,000 MB of memory, the Sales Silver Org vDC Resource Pool gets 125,000 MHz of CPU and 150,000 MB of memory to run their vApps.
So what's the big deal? If the Sales department over-commits their memory allocation which starts to cause memory ballooning or swapping it only affects the VMs in their resource pool. The same holds true with CPU, if the Sales department starts taxing the CPU and choking the virtual instances it will only affect the VMs in their resource pool.
Lets say all your tenants are in the root of the Silver Resource Pool. The CPU and memory resources will be used concurrently by multiple organizations. The developer of the Client Services vApp from the Sales Organization, Joe "The Maverick", decides to increase the compute resources from 2 vCPU to 16 vCPUs on 30 of his virtual machines at lunch time for stress testing. It will affect all the other tenants if he over-subscribes the CPU allocation on the Silver Resource Pool.
That makes the other tenants mad, your boss mad, his boss mad, and really makes for a bad day at the office! Protect yourself against Joe "The Maverick".