Contention scores for compute node per pCPU, per host, per thread

Every physical processor has a common cache called the last level cache. When VMS are allocated to CPUs they share the last level cache. Application inside VMs may contend for cache. Similarly network channels are shared and VMs contend for network. A node, socket and thread level contention helps in identifying noisy neighbors and make scheduling decisions accordingly. Direct scheduling to sockets with low contention can be used as a filter. The architecture for the solution will contain:
1. Agent code to measure of usage of LLC per thread
2. report the data back Ceilometer or other mechanisms OR
3. store the data at the node available by conf files variables
4. Creation of contention scores and re port the data back via Ceilometer OR
5. store the data at the node available by conf files variables