Robust Leasing for Virtual Infrastructure

Lease contracts are a powerful and general abstraction for negotiating and arbitrating control of shared networked resources. This paper addresses failure handling and recovery for leasing protocols and services. We present the design and implementation of a robust cooperative leasing service and illustrate its use for managing shared virtual clusters. The state update model is based on interacting recoverable state machines, an alternative to distributed transactions in which each participant recovers from a local log and then completes a selfsynchronization protocol with other participants to restore each lease to a globally consistent state. We explore the impact of hidden component dependencies in virtual infrastructure, and the role of the leasing core in responding to failures involving plug-in extensions: policy modules and configuration handlers for hosted services and resource components.