An Empirical Study of Segmented Linear Regression Search in LevelDB

The purpose of this paper is proposing a novel search mechanism, called SLR (Segmented Linear Regression) search, based on the concept learned index. It motivated by our observation that lot big data, collected and used previous studies, have linearity property, meaning keys their stored locations show strong linear correlation. This leads us to design where we apply segmentation into well-known machine learning algorithm, regression, for identifying location from given key. We devise two techniques, equal-size error-aware, with consideration both prediction accuracy overhead. implement proposal in LevelDB, Google’s key-value store, verify it can improve performance up 12.7%. In addition, find technique provides efficiency training while error-aware one tolerable noisy data.