In this demo, we attempt to map a number of services with a set of resources, where services are applications run on a data-center, and the resources are machines where they will run. Each service has resource requirements like CPU and Memory required, and each machine has certain resources to offer, as shown by the picture below. Essentially, we have a multidimensional bin-packing problem.
In Chapter 4 of Optimization and Mathematical Modeling in Computer Architecture, we discuss four different scenarios and formulations of this problem, including having heterogeneous hardware (Static Server Allocation Problem — SSAP), homogeneous hardware (Warehouse Server Allocation Problem — WSAP), time-varying service requirements (Time-Varying Server Allocation Problem — TSAP), and memory-interference sensitive allocation (Interference-Sensitive Server Allocation Problem — ISAP). The formulations are described further in the book.
Case Study 2:
Data-center resource managers are centralized software components which manage the execution of a large number of services. Co-locating multiple services on a machine could degrade performance, and hence it is critical for the resource manager to utilize information upon services and machines in effectively allocating machine resources.
In ILP, the problem is to statically determine the best co-locations of services on servers for resource requirements and service-level agreements to be satisfied.
Select from the following resource management models (mouse over for description).