Ravi A. Shankar

DynaPlan: Resource placement for application-level clustering

Creating a reliable computing environment from an unreliable infrastructure is a common challenge. Application-Level High Availability (HA) clustering addresses this problem by relocating and restarting applications when failures are detected. Current methods of determining the relocation target(s) of an application are rudimentary in that they do not take into account the myriad factors that influence an optimal placement. This paper presents DynaPlan, a method that improves the quality of failover planning by allowing the expression of a wide and extensible range of considerations, such as multidimensional resource consumption and availability, architectural compatibility, security constraints, location constraints, and policy considerations, such as energy-favoring versus performance-favoring. DynaPlan has been implemented by extending the IBM PowerHA clustering solution running on a group of IBM System P servers. In this paper, we describe the design, implementation, and preliminary performance evaluation of DynaPlan.