R-Opus: A Composite Framework for Application Performability and QoS in Shared Resource Pools

Abstract

We consider shared resource pool management taking into account per-application quality of service (QoS) requirements and server failures. Application QoS requirements are defined by complementary specifications for acceptable and time-limited degraded performance. Furthermore, a requirement specification is provided for both the normal case and for the case where an application server fails in the pool. Independently, the resource pool operator provides a resource access QoS commitment for two classes of service (CoS). These govern statistical multiplexing within the pool. A QoS translation automatically maps application demands onto the resource pool's CoS to best enable sharing. A workload placement service consolidates applications to a small number of servers while satisfying application QoS requirements. The service reports whether a spare server is needed or how applications affected by a single failure can operate according to failure QoS constraints using remaining servers until the failure can be repaired. A case study demonstrates the approach