A Characterization Study of NCSTRL Distributed Searching
Dushay, Naomi; French, James C.; Lagoze, Carl
NCSTRL, the Networked Computer Science Technical Reference Library, is a federated digital library based on the Dienst architecture. One aspect of this architecture is distributed searching, with digital library queries being dispatched from query routers to globally distributed indexers that process them and return results. We studied user data for a two-month period at five query routers in order to characterize some key performance aspects of distributed searching in an operational digital library. This study uncovered the following characteristics. Query processing at NCSTRL servers involves significant time waiting for responses from indexers. Each indexer's availability and response times appear unique to each query router. Different indexers' availability and response times are not similar from the viewpoint of a single query router. Query router waiting time for indexers is larger than indexer processing time, implying that communication time over the network is significant. We close by examining the breakdown of NCSTRL queries: the number of fielded vs. non-fielded queries, and the complexity of these queries.
computer science; technical report
Previously Published As