Title: Making XML Database Systems Scalable to Computer Resources and Data Volumes

Abstract: 
Increasing use of XML has emphasized the need for scalable database systems 
that are capable of handling a large amount of XML data efficiently. This study 
explores effective methods for making a scalable XML database system in the 
following aspects: (a) scalability to data volumes, (b) scalable XML processing 
with a shared-nothing PC cluster, and (c) scalable database processing on 
shared-memory multiprocessors. In the study of (a), we propose an XQuery 
processing scheme in which an XML document is internally represented as a set 
of blocks and can directly be stored on secondary storage. Our experimental 
results showed that our storage scheme is scalable to data volumes and 
outperforms competing schemes with respect to I/O intensive workloads. In (b), 
we discuss on-the-fly XML processing using shared-nothing PC clusters. We 
propose a scheme for distributed and parallel query processing that employs a 
pass-by-reference semantics by using remote proxy. Previously proposed methods 
that use pass-by-value semantics have often suffered from redundant 
communication between processor elements and limited inter-operator 
parallelism. To cope with these problems, we developed a distributed XML query 
processing scheme that leverages the benefit of lazy evaluation. Our 
experimental results showed that our proposed scheme obtains up to 22x speedups 
compared with competitive methods, and demonstrated the importance of 
distributed XML database systems to take pass-by-reference semantics into 
consideration. In (c), we explain the internal locking in the buffer management 
module that prevents databases from being scalable to the number of processors. 
We further propose a scalable buffer management scheme that employs 
non-blocking synchronization instead of locking-based ones. Our experimental 
results revealed that our scheme can obtain nearly linear scalability to 
processors up to 64 processors, although the existing locking-based schemes do 
not scale beyond 16 processors. Finally, we conclude our studies with examining 
our XML native database system built on top of the three contributions.