60 likes | 147 Views
Explore a policy model with hierarchical attributes and learn how to efficiently partition the hierarchy using top-down and bottom-up approaches. Discover solutions to clustering subjects and mapping them to servers based on data similarity.
E N D
Research Presentation ParthNagarkar Feb. 24, 2012
Background • We have attributes with hierarchical structure • We have defined a policy model • ok(subject_name, attribute_name, object_name) • E.g. ok (Mithila, Location, PHX) ok (Mithila, Location, California) U.S California Arizona PHX L.A. SFO Database Entries for an attribute “Location”
Motivation • In the previous example, we only had one subject. What if there are policies for more than one subject? • Why do we need to partition the hierarchy? • Types of wasted work: • Refinement work (root node) • Replacement work (leaf nodes) U.S Jung’s level CA AZ Xilun’s level PHX L.A. SFO Mithila’s level
Proposed Approach • For a cutoff/level in a hierarchy,
Proposed Approach (contd.) • Top-Down approach • Bottom-Up approach • Binary Search approach
More problems to solve • How to cluster subjects and map them accordingly to different servers? • Subjects that read similar data should be clustered together • Extend the model to multiple attributes • Implementation: • Given a cutoff, how to figure out if the data above the cutoff actually fits in the memory? • Finish the three approaches