Cache Design of SSD-Based Search Engine Architectures

Abstract

Caching is an important optimization in search engine architectures. Existing caching techniques for search engine optimization are mostly biased towards the reduction of random accesses to disks, because random accesses are known to be much more expensive than sequential accesses in traditional magnetic hard disk drive (HDD). Recently, solid-state drive (SSD) has emerged as a new kind of secondary storage medium, and some search engines like Baidu have already used SSD to completely replace HDD in their infrastructure. One notable property of SSD is that its random access latency is comparable to its sequential access latency. Therefore, the use of SSDs to replace HDDs in a search engine infrastructure may void the cache management of existing search engines. In this article, we carry out a series of empirical experiments to study the impact of SSD on search engine cache management. Based on the results, we give insights to practitioners and researchers on how to adapt the infrastructure and caching policies for SSD-based search engines.

blockNDP

“…In-storage NDP. In-storage NDP has been widely studied in various use-cases, such as databases [17,19,31,37], map-reduce [23,24,29], regex [8,28], searching [19,35,36], etc. A common theme of such works is to explore the idea of pushing computation to the storage devices for improved performance.…”

Boosting Search Performance Using Query Variations

“…List caching involves storing commonly accessed postings lists in fast-access memory [18,72]. For example, if the postings lists making up the index are stored on a SSD, a list caching strategy may opt to keep to n most accessed postings in main-memory.…”

“…The performance of all of the approaches degrades as ϵ increases, and they are vulnerable to inaccuracies if the clustering error might be high. The key point is that if a search engine receives an exact match well-known query, it can simply return a cached result, which is what is commonly done in practice [3,18,32,36,50,72]; what we are exploring here is the consequence of also allowing approximated matches to be exploited.…”

“…Caching is a simple and common approach for increasing query throughput at the cost of additional space consumption. Caches can be deployed at many levels of storage, including in-memory or on-disk [3,72]. In IR there are two major approaches to caching.…”

“…Alternatively, result caching involves storing a query along with the relevant results (or some proxy thereof) that were returned for the query [32,36]. In practice, both list and result caching are useful, and are deployed in tandem [3,72]. Most caches utilize historical data such as static query logs or sliding windows of recent queries to build models of what to cache, and when to cache it, and can also be personalized on a per-user basis [50].…”