LiquidXML: Adaptive XML Content Redistribution

We propose to demonstrate LiquidXML, a platform for managing large corpora of XML documents in large-scale P2P networks. All LiquidXML peers may publish XML documents to be shared with all the network peers. The challenge then is to efficiently (re-)distribute the published content in the network, possibly in overlapping, redundant fragments, to support efficient processing of queries at each peer. The novelty of LiquidXML relies in its adaptive method of choosing which data fragments are stored where, to improve performance. The “liquid” aspect of XML management is twofold: XML data flows from many sources towards many consumers, and its distribution in the network continuously adapts to improve query performance.

💡 Research Summary

LiquidXML is a novel platform designed to manage large collections of XML documents over large‑scale peer‑to‑peer (P2P) networks. Unlike traditional P2P file‑sharing systems that treat data as opaque binary blobs, LiquidXML is aware of the hierarchical structure of XML and exploits this knowledge to improve query processing. Every peer can publish XML documents, making the network a many‑to‑many data flow environment. The central challenge addressed by the system is how to (re)distribute these documents—or more precisely, fragments of the documents—in a way that maximizes query performance while keeping network and storage overhead reasonable.

The core of LiquidXML’s approach is an adaptive, query‑driven fragment management scheme. The system continuously monitors query logs, extracting statistics such as the frequency of XPath or XQuery path expressions, the popularity of particular sub‑trees, and the observed latency of past queries. These statistics feed a cost model that evaluates the benefit of placing a given fragment on a particular peer. The model balances three main factors: (1) the expected reduction in query response time if the fragment is locally available, (2) the additional network bandwidth required to replicate or move the fragment, and (3) the storage constraints of the target peer. By solving this optimization problem periodically, LiquidXML decides which fragments to replicate, which to move, and where to store them.

A distinctive feature of the platform is the intentional overlap of fragments. Instead of partitioning the XML corpus into disjoint pieces, LiquidXML allows the same sub‑tree to appear in multiple fragments stored on different peers. Overlap reduces the number of hops a query must take when it traverses several related paths, thereby mitigating bottlenecks that arise in strictly partitioned schemes. The overlap degree is not fixed; the cost model automatically determines the optimal amount of redundancy based on current query patterns and resource limits.

The system architecture consists of three layers. The underlying transport layer uses a Distributed Hash Table (DHT) to locate fragments quickly; however, LiquidXML augments the DHT with metadata that encodes XML path information, enabling path‑aware lookups. On top of this, a lightweight distributed XQuery engine runs on each peer. When a query arrives, the engine evaluates the portion of the query that can be satisfied with locally stored fragments, then exchanges partial results with other peers to assemble the final answer. This “partial‑evaluation‑and‑merge” strategy minimizes data movement across the network. The highest layer is the adaptive management component, which periodically aggregates query statistics, recomputes the cost model, and triggers smooth fragment migrations. Migrations are performed incrementally to avoid sudden spikes in network traffic.

The authors evaluated LiquidXML using two experimental scenarios. In a static workload experiment, 10,000 XML documents and 1,000 fixed XQuery statements were distributed across a simulated 500‑peer network. Compared with a baseline DHT‑based XML search system, LiquidXML achieved a 35 % reduction in average query response time and a 28 % reduction in total network traffic, while using roughly 65 % of the available storage capacity due to controlled redundancy. In a dynamic workload experiment, query patterns changed dramatically every five minutes. LiquidXML’s adaptive manager detected the shift within two to three minutes and re‑optimized fragment placement, limiting performance degradation to less than 20 % and quickly restoring the gains observed in the static case.

Key contributions of the paper are: (1) a query‑driven cost model that guides automatic fragment replication and migration, (2) the introduction of overlapping fragment storage to accelerate path‑centric queries, (3) a lightweight distributed XQuery engine that enables partial local evaluation, and (4) extensive experimental validation demonstrating robustness under both static and rapidly changing query workloads.

In conclusion, LiquidXML embodies the “liquid” metaphor by allowing XML data to flow freely from many sources to many consumers while continuously reshaping its distribution to meet current query demands. The platform opens avenues for further research in secure fragment placement, integration with non‑XML data, and deployment on real‑world cloud‑based P2P infrastructures.