Polus: Growing Storage QoS Management beyond a “Four-year Old Kid”

Abstract

Policy-based storage management has been advertised as the silver bullet to overcome the complexity that limits the amount of storage that can be managed by system administrators. Key to this approach are: a mechanism to specify quality of service (QoS) goals; a canonical virtual model of storage devices and operations; and the mapping of the high level QoS goals to low level storage device actions. In spite of prior research and industrial standards the latter problem results in complex, manual, error-prone processes that burden system administrators and prevent the widespread acceptance of policy-based storage management. This paper proposes the Polus framework which specifically addresses this open problem. Polus removes the need for system administrators to write code that maps the QoS goals to low level system actions. Instead, it generates this mapping code by using a combination of rule-of-thumb specification mechanism, a reasoning engine and a learning engine to change the implementation paradigm of policy-based storage management. This paper also provides a quantitative analysis of the Polus framework within the context of a storage area network (SAN) file system to verify the feasibility of this new approach.