Merging and comparison of RLS- and DHT-based resource indexing and management

This is a project proposal (20 credits) for one or two master students. The projects will be carried out at FOI, the Swedish Defence Research Agency at the department of Systems Modelling. The project is expected to start during the summer or fall of 2005.

Keywords

Resource Indexing, Resource Management, Grids, Globus Toolkit, Replica Location Service, RLS, Distributed Hash Tables, DHT, Distributed systems

Background

The NetSim project at FOI is a project aimed at network based modelling and simulation. The vision is to improve the M&S-process and make it more efficient in order to support simulation models throughout their entire life cycle. A part of this work is to create M&S-related services for the Network Based Defence. Therefore we are building a framework for the development, execution, use and reuse of distributed simulation Models. A part of this framework is the Distributed Repository, DR, a resource library.

The purpose of DR is to use and share different kinds of resources independent of their location, owner or organization. Resources in this case can be anything from employees, files and simulations models to available execution power. The main functionality in DR is, besides storage and downloading of resources also description, removal and semantic search of resources. A prototype implementation of DR has been done based on techniques and methods related to the Semantic Web and Grids. Examples of used techniques and programs are Globus Toolkit, Grid Services, PostgreSQL and the implementation has mainly been done in Java.

During 2005 the work with DR will continue by further development of certain parts and by connecting it to other parts of the NetSim-environment. The following master projects are a part of the ongoing work with DR and NetSim.

Proposal

Replica Location Service, RLS, was developed as a part of the Globus Toolkit as a means of keeping track of one or more copies, or replicas, of files in a Grid environment. RLS is a distributed registry as it consists of different servers at different sites. RLS maintains associations, or mappings between logical file names and one or more physical file names of replicas.

Distributed Hash Tables, DHT, are algorithms for mapping keys to values in decentralized, scalable, self configuring and self stabilizing infrastructures. These infrastructures can then be used to support more complex services such as indexing and mapping of resources and their names.

The scope of the master project is to study DHTs and RLS, migrate the existing implementation of RLS in DR from Globus Toolkit 3.2 to Globus Toolkit 4 and implement a merge of RLS and a suitable DHT. Analyze the result and write a report in English. Some of the interesting questions to answer are: What are their advantages or disadvantages? Which is faster and more reliable in a situation like the DR? Is the combination good for the Distributed Repository?

Prerequisites

- The students must understand basic concepts and methods in distributed systems and Grid computing.
- Have good programming skills in Java.
- Preferably have some knowledge in Modeling and Simulation, Distributed Algorithms and Distributed Hash Tables, DHT.

References

[1] Sima Baymani, Emil Stridfeldt, Carbonara ¿ a semantically searchable distributed repository. Master of Science Thesis, Royal Institute of Technology Stockholm, Sweden 2005, IMIT/LECS-2005-44, http://web.it.kth.se/~rassul/exjobb/rapporter/sima-emil.pdf
[2] Globus Alliance and Globus toolkit, http://www.globus.org/
[3] RLS, http://www.globus.org/toolkit/data/rls/
[4] DHT, http://en.wikipedia.org/wiki/Distributed_hash_table (see also Chord and Pastry)
[5] Min Cai, Ann Chervenak, Martin Frank, A Peer-to-Peer Replica Location Service Based on A Distributed Hash Table, 2004 http://www.globus.org/alliance/publications/papers/sc2004v15.pdf