The Cambridge eScience Centre

 

FutureGrid




A program for long-term research into Grid systems architecture.

Project Overview (PDF)

More information is available on the Futuregrid project page at the University of Cambridge Computer Laboratory, including a list of the project staff.

Conference paper: FutureGRID : A Program for long-term research into GRID systems architecture

Flexible server discovery

How Search Time Varies with Number of Items Future grid systems will federate large numbers of computational nodes drawn from a wide variety of organizations. Existing server discovery mechanisms allow clients to select servers based on system properties -- operating system, hardware kind, specialist facilities and so on. However, a large class of clients impose requirements based primarily on server location -- for example to place an OGSA service close to a potential user community, or to spread critical replicas widely throughout the network. The same problems are seen today in PlanetLab-style network testbeds, or in public computing platforms such as in our XenoServers project.

XenoSearch is an advanced server discovery system which we have built as part of the FutureGrid project. Clients use it to discover the most suitable server on which to deploy a service. Queries specify the desired network location for the service, along with the kind of environment required and the QoS attributed desired. For instance, we can search for servers which are near to a particular existing machine (e.g. for fast interaction), or which are equidistant between a set of clients (e.g. for multi-user interaction).

Example operation pictureXenoSearch receives advertisements from servers and maps them to points in a multidimensional space. Different dimensions represent different attributes -- topological location, resource availability and so on. XenoSearch then uses a novel scheme to summarize the advertisements falling within each range of each dimension. Ranges can then be queried and combined to obtain a summary of an arbitrary range. This is built over a distributed hashtable, allowing XenoSearch itself to be provided by a distributed service.

XenoSearch was described in our paper at HPDC-12/GGF-8. In ongoing work we are investigating mechanisms for describing more complex server requirements -- in particular searches requesting a set of inter-related services.

David Spence
David.Spence@cl.cam.ac.uk
University of Cambridge Computer Laboratory

  



Comments to the webmaster. Contact: Tel (+44/0) 1223 764282, Fax (+44/0) 1223 765900