Distributed Cache Table: Efficient Query-Driven Processing of Multi-Term Queries in P2P Networks

The state-of-the-art techniques for processing multi-term queries in P2P environments are query flooding and inverted list intersection. However, it has been shown that due to scalability reasons both methods fail to support full-text search in large scale document collections distributed among the nodes in a P2P network. Although a number of optimizations have been suggested recently based on the aforementioned techniques, little evidence is given on their scalability. In this paper we suggest a novel query-driven indexing strategy which generates and maintains only those index entries that are actually used for query processing. In our approach called Distributed Cache Table (DCT), by analogy with Distributed Hash Table (DHT), we suggest to abandon the difference between data indexing and query caching, and to store result sets (caches) for the most profitable queries. DCT employs a distributed index to efficiently locate caches that can answer a given multi-term query and broadcasts the query to all the peers only if no such caches were found. Evaluations on real data and query loads show that DCT converges to a high cache-hit ratio and indeed offers a large-scale distributed solution for storing and efficient querying of vast amounts of documents in the P2P setting. DCT achieves two orders of magnitude improvement in traffic consumption compared to a standard distributed single-term indexing approach.

Karl Aberer, Gleb Skobeltsyn

2006

Query-driven indexing in large-scale distributed systems

Gleb Skobeltsyn

Efficient and effective search in large-scale data repositories requires complex indexing solutions deployed on a large number of servers. Web search engines such as Google and Yahoo! already rely upon complex systems to be able to return relevant query results and keep processing times within the comfortable sub-second limit. Nevertheless, the exponential growth of the amount of content on the Web poses serious challenges with respect to scalability. Coping with these challenges requires novel indexing solutions that not only remain scalable but also preserve the search accuracy. In this thesis we introduce and explore the concept of query-driven indexing – an index construction strategy that uses caching techniques to adapt to the querying patterns expressed by users. We suggest to abandon the strict difference between indexing and caching, and to build a distributed indexing structure, or a distributed cache, such that it is optimized for the current query load. Our experimental and theoretical analysis shows that employing query-driven indexing is especially beneficial when the content is (geographically) distributed in a Peer-to-Peer network. In such a setting extensive bandwidth consumption has been identified as one of the major obstacles for efficient large-scale search. Our indexing mechanisms combat this problem by maintaining the query popularity statistics and by indexing (caching) intermediate query results that are requested frequently. We present several indexing strategies for processing multi-keyword and XPath queries over distributed collections of textual and XML documents respectively. Experimental evaluations show significant overall traffic reduction compared to the state-of-the-art approaches. We also study possible query-driven optimizations for Web search engine architectures. Contrary to the Peer-to-Peer setting, Web search engines use centralized caching of query results to reduce the processing load on the main index. We analyze real search engine query logs and show that the changes in query traffic that such a results cache induces fundamentally affect indexing performance. In particular, we study its impact on index pruning efficiency. We show that combination of both techniques enables efficient reduction of the query processing costs and thus is practical to use in Web search engines.

EPFL2009

HPCache: Memory-Efficient OLAP Through Proportional Caching

Anastasia Ailamaki, Periklis Chrysogelos, Hamish Mcniece Hill Nicholson

Analytical engines rely on in-memory caching to avoid disk accesses and provide timely responses by keeping the most frequently accessed data in memory. Purely frequency- & time-based caching decisions, however, are a proxy of the expected query execution speedup only when disk accesses are significantly slower than in-memory query processing. On the other hand, fast storage offers loading times that approach or even outperform fully in-memory query execution response times, rendering purely frequency-based statistics incapable of capturing impact of a caching decision on query execution. For example, caching the input of a frequent query that spends most of its time processing joins is less beneficial than caching a page for a slightly less frequent but scan-heavy query. As a result, existing caching policies waste valuable memory space to cache input data that offer little-to-no acceleration for analytics. This paper proposes HPCache, a buffer management policy that enables fast analytics on high-bandwidth storage by efficiently using the available in-memory space. HPCache caches data based on their speedup potential instead of relying on frequency-based statistics. We show that, with fast storage, the benefit of in-memory caching varies significantly across queries; therefore, we quantify the efficiency of caching decisions and formulate an optimization problem. We implement HPCache in Proteus and show that i) estimating speedup potential improves memory space utilization, and ii) simple runtime statistics suffice to infer speedup expectations. We show that HPCache achieves up to 12% faster query execution over state-of-the-art caching policies, or 75% less in-memory cache footprint without deteriorating query performance. Overall, HPCache enables efficient use of the in-memory space for input caching in the presence of fast storage, without any requirement for workload predictions.