Introduction

Managing shared cyberinfrastructure resources is a fundamental challenge for service hosting and utility computing environments, as well as the next generation of network testbeds and grids. This paper investigates an approach to networked resource sharing based on the foundational abstraction of resource leasing.

We present the design and implementation of Shirako, a toolkit for a brokered utility service architecture.¹Shirako is based on a common, extensible resource leasing abstraction that can meet the evolving needs of several strains of systems for networked resource sharing--whether the resources are held in common by a community of shareholders, offered as a commercial hosting service to paying customers, or contributed in a reciprocal fashion by self-interested peers. The Shirako architecture reflects several objectives:

• Autonomous providers. A provider is any administrative authority that controls resources; we refer to providers as sites. Sites may contribute resources to the system on a temporary basis, and retain ultimate control over their resources.

• Adaptive guest applications. The clients of the leasing services are hosted application environments and managers acting on their behalf. We refer to these as guests. Guests use programmatic lease service interfaces to acquire resources, monitor their status, and adapt to the dynamics of resource competition or changing demand (e.g., flash crowds).

• Pluggable resource types. The leased infrastructure includes edge resources such as servers and storage, and may also include resources within the network itself. Both the owning site and the guest supply type-specific configuration actions for each resource; these execute in sequence to setup or tear down resources for use by the guest, guided by configuration properties specified by both parties.

• Brokering. Sites delegate limited power to allocate their resource offerings--possibly on a temporary basis--by registering their offerings with one or more brokers. Brokers export a service interface for guests to acquire resources of multiple types and from multiple providers.

• Extensible allocation policies. The dynamic assignment of resources to guests emerges from the interaction of policies in the guests, sites, and brokers. Shirako defines interfaces for resource policy modules at each of the policy decision points.

Section 2 gives an overview of the Shirako leasing services, and an example site manager for on-demand cluster sites. Section 3 describes the key elements of the system design: generic property sets to describe resources and guide their configuration, scriptable configuration actions, support for lease extends with resource flexing, and abstractions for grouping related leases. Section 4 summarizes the implementation, and Section 5 presents experimental results from the prototype. The experiments evaluate the overhead of the leasing mechanisms and the use of leases to adapt to changes in demand. Section 6 sets Shirako in context with related work.