Community Services
Overview
The current generation of Grid computing is focusing on how to "elevate" the level of abstraction to better enable Grid application designers and end-users to solve "real problems". The emergence of service-oriented architectures such as OGSA/OGSI are evidence of this trend. This model will succeed only if it is easy to deploy and manage services, users find that the services meet their specific needs, and the services themselves do not compromise Grid-wide policies.
The Community Services project is addressing these issues for high-end services, that is services which are resource-intensive in terms of computation, memory, or data storage. These services typically encapsulate parallel and distributed computing. The central challenge is that when users and services come together on the Grid, adaptation is required on many levels. Services must be adaptive to users because popular services can expect multiple concurrent users which will require adaptive resource management in the form of resource multiplexing and service replication to maintain scalable levels of performance. The Grid also induces a need for service adaptivity due its dynamics - resource fluctuation and shared ownership, failure, heterogeneity in terms of resources, platforms, and infrastructure. Current practice leaves the problem of adaptation to the service provider (or perhaps the user). However, embedding adaptive code within services is an arduous task for the service provider.
To address this problem, this project is developing re-usable middleware that simplifies this task for the service provider. Our goal is to develop flexible techniques and re-usable middleware in support of adaptive services that satisfies users, service providers, and the Grid community, and to validate this middleware by applying it to DOE-relevant services in Grid testbeds. We propose to encapsulate this middleware within several adaptive service classes: adaptive resource providers (ARP), and adaptive Grid services (AGS). The middleware will consist of performance prediction (PP), resource management (RM), and ARP selection (AS) modules. Several classes of adaptive Grid services will be investigated including data-intensive HENP-relevant services.[Contact Us] [U of M DCSG] [DOE National Collaboratories]
Distributed Computing Systems Group
Dept. of Computer Science & Engineering
University of Minnesota
Last updated : July 5. 2003