Relationally Mapping XML Queries For Scalable XML Search

Abstract

The growing trend of using XML to share security In Fig. 1, we show the total query time for each query. Also data requires scalable technology to effectively manage the shown are the mean and total times for each run. All timings volume and variety of data. Although a wide variety of methods given represent the average execution time of the queries (in exist for storing and searching XML, the two most common gvn rere the average exeut e of the q eries techniques are conventional tree-based approaches and relational random order) over five runs. To ensure a cold cache, the server approaches. Tree-based approaches represent XML as a tree was rebooted between runs. Overall, the relational approach and use indexes and path join algorithms to process queries. outperformed the tree-based approach on all five collections. In contrast, the relational approach seeks to utilize the power However, the tree based approach outperformed the relational of a mature relational database to store and search XML. This approach for both quantifier queries. On average, the relational method relationally maps XML queries to SQL and reconstructs approach took 17.5 times as long to execute the 8GB queries the XML from the database results.We use the XBench benchmark to compare the scalability of than the 500MB queries. This is very close to the expected the SQLGenerator, our relational approach, with eXist, a popular linear scaling factor of 16. On the other hand, the tree based tree-based approach.approach took 55.6 times as long to execute the queries on I. INTRODUCTION the 8GB collection. XML is a flexible and powerful tool that enables vital Collection Size security sharing in heterogeneous environments [1]. Since 500MB 1GB 2GB 4GB 8GB tree rel tree rel tree rel tree rel tree rel XML can be extended to include domain specific tags, inQl 15.52 0.43 26.75 0.44 59.41 0.51 131.11 0.71 484.24 0.67 formation can be encoded with meaningful structure and Q5 t4.52 0.77 25.72 0.78 56.20 0.96 129.61 1.27 498.83 1.47 Q6 2.03 31.08 4.64 50.26 t8.97 52.37 28.73 183.10 TIMEOUT 466.58 semantics that allow rapid information sharing among devices Q7 2.85 25.0 6.43 38.88 10.32 71.70 26.94 243.15 217.72 594.19 and organizations. We examine the conventional tree approach Q8 14.57 0.58 25.89 0.51 56.52 0.69 131.00 0.87 478.43 1.10 Q9 14.73 0.40 25.38 0.41 53.50 0.46 124.22 0.63 426.58 0.72 and the relational mapping of XML queries to determine which QIO 16.22 1.16 29.89 1.51 68.35 2.73 139.53 5.97 1,850.99 13.22 method has the potential to search large collections of XML 100.71 projects/xbench/index.html) to create a heterogeneous total 158.28 63.09 278.09 96.94 591.7 151.88 1741.33 477.44 8001.86 1107.84collection of multiple schema data-centric XML documents. We generated an 8GB collection from the modified XBench templates. The 500MB, 1GB, 2GB, and 4GB collections were created from random subsets of the 1GB, 2GB, 4GB, From the timing results presented, we conclude that the and 8GB collections respectively. relational approach is highly scalable to i...

Using a relational database for scalable XML search

“…Although the conclusion of this study showed linear execution time for eXist, the collections used were not large enough to draw reliable conclusions about the scalability of eXist on modern, real-world collections. Similarly, the SQLGenerator performed initial scalability studies by running a wide range of XML queries on increasingly large collections ranging in size from 500 MB to 8 GB [15].…”