Performance Management for Large Scale Service Delivery Platforms

Managing performance of modern Service Delivery Platforms (SDPs) is a challenging task due to scale, dynamicity and heterogeneity of these systems. With the presence of a plethora of widely divergent applications, it becomes almost impossible to model these SDPs accurately enough for basic management tasks. The dynamicity of these platforms adds another dimension of complexity. Precisely due to these reasons, traditional learning based approaches are not adequate since these approaches are inherently very slow. In this paper, we present a novel technique for performance management in SDP environment to overcome these challenges. In our approach, we construct a multi-dimensional performance model grid called datagrid. Datagrid presents a unified management approach for various system management tasks such as provisioning, fault management and SLA management. According to our understanding, ours is a first system to have a unified approach for large scale SDP environments for multiple management tasks. We have implemented our technique in real distributed system and it shows promising results.