Performance Improvements for Search Systems Using an Integrated Cache of Lists+Intersections

Abstract

Modern information retrieval systems use sophisticated techniques for e ciency and scalability purposes. Among the most frequent such techniques is the implementation of several levels of caching. The main goal of a cache is to speedup computation by exploiting frequent, recent or costly data used in the past. In this study we propose and evaluate a static cache that works simultaneously as list and intersection cache, o↵ering a more e cient way of handling cache space. In addition, we propose e↵ective strategies to select the term pairs that should populate the cache. Simulation using two datasets and a real query log reveal that the proposed approach improves overall performance in terms of total processing time, achieving savings of up to 40% in the best case.

The Impact of Using Combinatorial Optimisation for Static Caching of Posting Lists

“…For query logs simulated using the method in Section 4, we cannot control repeated queries. Experimental settings We experiment with cache sizes of 200, 600 and 1000 MB (cache sizes can vary between 100 MB to 16 GB [14]) and fix the size of a posting to 8 bytes. We use ClueWeb09 cat.…”

“…the frequency of t in each document [6]. Posting list caching can reduce the amount of disk I/O involved [13,14] in query processing, affords higher cache utilisation and hit rates than result caching [12], and can combine terms to answer incoming queries. Our contribution: We show that static caching of posting lists can be modelled in a principled manner using constrained combinatorial optimisation (CCO), a standard method that has yielded great improvements in many fields [9,Chap. 35], and we provide a principled investigation of whether CCO would yield better solutions (preferably using modest extra computational resources) than greedy methods.…”

“…Much prior work has been devoted to caching posting lists [3,4,5,6,10,11,13,14,15]. Zhang et al [15] benchmark five posting list caching policies and find LFU (least frequently used -cache members are evicted based on their infrequency of access) to be superior, and that cache hit rates for static posting list are similar to the LFU, but with less computational overhead.…”

“…Zhang et al [15] benchmark five posting list caching policies and find LFU (least frequently used -cache members are evicted based on their infrequency of access) to be superior, and that cache hit rates for static posting list are similar to the LFU, but with less computational overhead. An integrated cache that merges posting lists of frequently co-occurring terms to build new posting lists in the inverted index is used by Tolosa et al [14]. Using a cost function that combines disk lookup and CPU time, the integrated cache improves performance over standard posting list caching by up to 40%.…”

A machine learning approach for result caching in web search engines

“…Li, Lee, Sivasubramaniam, and Giles (2007) propose a hybrid architecture involving result, posting list, and document caches. Tolosa, Becchetti, Feuerstein, and Marchetti-Spaccalema (2014) present a technique that couples posting lists and term intersection lists in a single static cache. Zhou, Li, Dong, Xu, and Xiao (2015) also propose a three-level cache hierarchy consisting of a result cache, posting list cache, and a term intersection cache, generated using the top-N frequent item sets (terms) mined from query logs.…”