Title: AReNA: Adaptive Distributed Catalog Infrastructure Based On Relevance Networks
1AReNA Adaptive Distributed Catalog
Infrastructure Based On Relevance Networks
- Vladimir Zadorozhny, University of Pittsburgh,
Pittsburgh, PA - Avigdor Gal, Technion, Haifa
- Louiqa Raschid, University of Maryland, College
Park, MD - Quiang Ye, University of Pittsburgh, Pittsburgh,
PA
Nebula Project http//db.sis.pitt.edu/projects/Ne
bula
2Networked Query Processing
Relevant statistics response time, network
delay, data transfer rate, etc.
Statistics about data
Data sources are remote, distributed,
heterogeneous
3Networked Queries with Distributed Catalog
Statistics about data
data
Scalability ?
4Performance Monitoring for Server Selection
- Objective maintaining comprehensive performance
repository for WANs (e.g., access latencies).
Motivated, in part, by the evolution of
information-centric name resolution services,
e.g., CNRI Handle system.
Handle system
- Challenge scaling to the presence of hundreds of
servers and thousands of clients, managing
millions of constantly changing Performance
Profiles.
LEGEND
client
5Profile-Based Performance Monitoring
PM Aggregation ?
6Aggregated Latency Profiles
- A client/server pair is characterized by
Individual Latency Profiles (iLP). iLPs capture
latency distributions experienced by clients when
connecting to a server.
7Discovering Non-random Associationswith
Relevance Networks (RNs)
We adopt RNs as a management tool, to manage
large numbers of iLPs.
8Relevance Networks
9AReNA Architecture
AReNA dynamically analyzes and visualizes
meaningful relationships among client/ server
pairs using Relevance Networks (RNs).
Relationships are evaluated using passive
measurements made by client applications and
gathered on a continuous basis. RNs allow AReNA
managing thousands of constantly changing iLPs
Large-Scale Experimental Testbeds CNRI Handle
System PlanetLab Overlay Around 50 000 Latency
Profiles
10AReNA Screenshot
11 Demo
Tuesday 1600-1730
Friday 0900-1030