Tim Freeman from GRIDVM.org, along with R. Bradshaw, N. Desai and K. Keahey, wrote a paper about virtual appliance configuration and management that was accepted for the TeraGrid 2007 conference and they have made that paper available online.  The paper examines configuration and security issues that large and heterogeneous deployments of virtual appliances/workspaces will face.  Great job guys!

The use of virtualization in Grid computing has seen a lot of interest lately. However, while much effort has been expanded on developing the capabilities of Virtual Machine Monitors (VMMs) and associated tools and services relatively little has been done to investigate the requirements underlying the scalable production, deployment, and management of VM images. At the same time, the clear understanding of requirements and capabilities in this area is critical to creating progress in exploring the applications of virtualization. In this paper, we investigate the issues and propose some of the solutions relevant to this question.

Introduction

One of the main motivations for using virtual machines is to easily and scalably provide on-demand environments - a VM image can be prepared and finely customized ahead of time and then deployed or taken down in a matter of milliseconds. This in practice significantly expands the set of configurations a site can host as we can now switch between even very complex configurations without seriously impacting the time in which resources are available for computation. In practice however, the need to maintain a large number of VM images - potentially orders of magnitude more than physical resources - as well as the sheer volume required for the storage of such images would pose a barrier to the scalability of this approach.

Further, the ability to deploy a pre-configured image decouples the (typically long) environment configuration process from the (now short) process of binding environments to resources. A VM image can thus be easily and quickly deployed on any site that happens to have available resources. This enables the emergence of a new provisioning model in which a site does not need to understand in detail all the configurations required by its users. However, it poses other problems. First, although the bulk of configuration can be done ahead of time, a small but critical amount of configuration has to be carried out when an image is deployed. This includes for example assigning network addresses and adjusting the  configuration of applications relying on them, providing a host certificates for an image, pointing the appliance at site services, and generally making it aware of its deployment context. Further, the site administrator needs to be able to establish trust in an image - verify that the image configuration complies with site policies concerning for example offsite root access or presence of software with known security exploits. Without mechanisms addressing these issues, in practice only siteadapted images will be deployable in practice, with effects opposite to what is desired.

Read the entire paper, here.