Enterprise Data Center Globality Measurement

Measurement of the globality of On-Line Transaction Processing (OLTP) workloads in Enterprise Data Centers is considered. Providing OLTP workload isolation (application, services and databases) for performance-sensitive enterprise workloads, so that activity in one workload cannot interfere with another, remains a challenge. We demonstrate that traditional aggregate OLTP Workload globality measurement frameworks can generate mis-leading globality measures. We propose a higher-order globality measurement framework which addresses this problem. We derive two high dimensional structured measurement matrices, namely a template and measurand matrix, with special spectral properties, which account for globality measurement 1) boundedness, 2) programmability, 3) multiplicity, 4) relativity, 5) spatial correlation and 6) the appropriate sensitivity of the measure to changes in the distribution of the workload. We demonstrate that these properties are exhibited by the new measure by ordering OLTP workloads by their globality measure. We evaluate the measure using a stochastic layered block model for data center topology and OLTP workload generation and demonstrate that it is consistent.