There is wide agreement that an essential key to the new era of simulation-intensive, collaboration-enabled scientific computing, envisioned by DOE’s Scientific Discovery through Advanced Computing (SciDAC) initiative, is the development of advanced network and middleware services that can provide reliable, fast, flexible, scalable, and efficient delivery of data to support distributed and high performance applications of all types. The research on Logistical Networking that we propose represents an aggressive approach that aims to extract maximum leverage to achieve this end from the ongoing, exponential growth in all types of computing resources — processing power, communication bandwidth, and storage.

Logistical Networking is the global scheduling and optimization of data movement, storage and computation based on a model that takes into account all of the network’s underlying physical resources, especially storage. By adding a uniform model of storage to the definition of the network and exposing it for direct scheduling, Logistical Networking will enable SciDAC application designers to make much stronger assumptions about the ability of next generation applications to manage distributed state and engage in asynchronous communications of all types. We believe that sharing storage resources in this way is essential to robust collaboration based on data sharing that SciDAC envisions. The research proposed here will allow us to

• perfect the engineering of the necessary middleware for remote storage management that Logistical Networking requires, making it fast, robust, scalable, and easy to use,
• experiment with and explore the capabilities of this middleware in a range of innovative applications with high performance and/or unique functionality; and
• disseminate Logistical Networking by deploying these technologies and applications to a broad community of users on a special testbed made available for this purpose.

We have designed and implemented fundamental, low level middleware, called the Internet Backplane Protocol (IBP) for managing remote storage to support Logistical Networking. With a stable, production quality implementation of IBP in hand and significant experience using it successfully in a variety of experiments and applications, we are prepared not only to push our research into promising new areas, but also develop the software infrastructure necessary to build the storage-enabled Internet and make Logistical Networking ubiquitous within the SciDAC community. This work falls into three main parts.

First, we will develop the complementary middleware components necessary to make IBP more robust, scalable, and easy to use. The technologies include the exNode, which is an XML-encoded data structure for building a strong file abstraction on the weak semantic properties of IBP, and a Data Mover Interface for optimizing transfers between IBP "depots." This effort will lay an adequate foundation for large-scale deployment of storage-enabled internetworking across the national research community.

Second, with these software tools in hand, we will explore fertile new areas and applications of research. This will include investigations of the generalized logistical routing problem, Logistical Session Layer technology for performance optimization and dynamic overlay networks, application-independent caching services for state management in computationally intensive network applications, and universal IBP-mail that supports e-mail attachments of arbitrary size for distributed scientific collaboration.

Finally, we will continue disseminate this technology via a research testbed called the Logistical Backbone (L-Bone). This will put Logistical Networking technology directly into the hands of the research community accessible to Internet2, ESNet, and other advanced networks.

Year 1

• Continuing development of IBP depot
• Deployment of early L-Bone tools
• Initial development of exNode libraries
• IBP Mail available to users
• IBP applications demonstrated at SC’01
• exNode support in NetSolve
• Reliability/performance coscheduling alpha
• Simulation of allocation policy

Year 2

• Initial deployment of generalized caching infrastructure
• Initial deployment of LSL-based logistical overlay network on ESNet
• Development of wide-area logistical peering mechanisms and policies
• Resolution for highly volatile storage resources

.

Year 3

• Experimental IBP architectures
• Large scale measurement and simulations