``The Logistical Computing Stack -- A Design For Wide-Area, Scalable, Uninterruptible Computing''

James S. Plank and Micah Beck.

A four-page paper giving an overview of our presentation given at The Workshop on Scalable, Uninterruptible Computing, part of DSN 2002: Dependable Systems and Networks, Bethesda, Maryland, June, 2002.

This was published by IEEE press as a supplement to the conference proceedings.

Available via anonymous ftp to cs.utk.edu in pub/plank/papers/DSN-2002-SUC.ps and pub/plank/papers/DSN-2002-SUC.pdf and

Introduction

As the most recent wave of innovation in network computing crested near the end of the last decade, the enthusiasm for the next generation of information technologies was running very high. Inspired by the global impact of the Internet and the Web, and fueled by exponential increases in all basic computing resources, the network computing community spawned a range of ambitious new approaches and applications (e.g.~active networks, mobile agents, ubiquitous computing, scientific collaboratories, etc.), many of which received significant development and had impressive early deployments. What seems evident today, however, is that this group of much anticipated innovations has had relatively little impact on advances in information technology practice; despite the rising flood of resources, the next wave of innovation has not yet materialized.

The concept of computation and storage residing in the network, instead of just being connected by the network is powerful. An in-the-network computing framework driven by this concept and supported by a highly programmable interface to individual application developers will liberate applications to scale to millions of instances that intelligently utilize a rich amount of computational, storage and network bandwidth resources.

The current manner in which processor and storage resources are managed in networked computing does not measure up to this challenge. The crux of this challenge, as we see it, is to enable users to benefit from provisioning of resources that are external not only the user's workstation but also to their immediate organization. This is an important part of the power of infrastructures such as the Internet and the Web: users have access to resources that they have no awareness of until they need to use them.

A guiding idea of our research program in Logistical Computing and Internetworking (LoCI), and the work on Logistical Networking at its core, is that the chief impediment to progress in this area is the failure to rethink and redesign the way in which storage and processor resources are integrated into the underlying ``fabric layer'' of resources [Foster, Kesselman, Tuecke 2001], the ``programmatic substrate'' [Amir, McCanne and Katz 1998] on which these new technologies are trying to build. In this paper, we describe the co-design of the storage and processing elements of this fabric that will enable globally scalable and uninterruptible computing applications.

Postscript of the paper

PDF of the paper

Citation Information