ALSM represents a major new stage in a research program in Logistical Computing and Internetworking, (LoCI), for distributed state management. The IP model is too limited for distributed computing because it does not include the possibility of managing application state between end systems (i.e. at intermediate nodes). The unique contribution of the LoCI approach is that it not only models this possibility, it does so without sacrificing the scalability essential to networking.
While not viewing the IP model as sufficient for distributed systems, the project's strategy is to reapply the end-to-end design principles that underlie that model to storage and processor resources. The first target was storage. Following the end-to-end paradigm, the PIs created the Internet Backplane Protocol (IBP), which is a primitive mechanism for managing distributed storage. With IBP-based intermediate nodes, called depots, the network itself can be infused with storage resources that can be shared, scaled up, and exposed for external scheduling, much like IP service is. A testbed provisioned with more than 22TB of storage, spread among 250 depots in 20 countries has been deployed, and a variety of application groups are currently using the testbed to explore the technology.
Following the LoCI design philosophy, ALSM.s guiding hypothesis is this: If, by adhering to end-to-end principles, a generic, best effort computing service can be integrated into ascalable logistical network, then a programmable network resource fabric can be created that is capable of providing the kind of active state management that grid applications require.
The work will encompass the following main activities:
o Design Active Depots and ALSM middleware: will optimize active depot architecture to efficiently support data transformation; middleware will be created to map a high level network view onto the weak semantics of IBP and the NFU, using sophisticated end-to-end techniques to generate reliability, performance, etc. o Enable GridSolve-L application environment: NetSolve will be adapted to use ALSM middleware so that NetSolve applications can issue logistical directives to exploit data locality patterns and offload state processing to active depots in the ALSM infrastructure. The result of this integration is GridSolve-L. o Validate ALSM in a grid visualization application: We will investigate the use of ALSM to implement a remote visualization pipeline, using a real-world medical imaging application under GridSolve-L; operations will separate as appropriate onto active depots for state management and GridSolve-L for major computation.
