In-Memory Analytics

In-Memory Analytics

Jorge Manjarrez Sánchez (Instituto Tecnologico Superior de Jerez, Mexico)
DOI: 10.4018/978-1-5225-7598-6.ch028

Abstract

Analytics is the processing of data for information discovery. In-memory implementation of machine learning and statistical algorithms enable the fast processing of data for descriptive, diagnostic, predictive, and prescriptive analytics. In this chapter, the authors first present some concepts and challenges for fast analytics, then they discuss some of the most relevant proposals and data management structures for in-memory data analytics in centralized, parallel, and distributed settings. Finally, the authors offer further research directions and some concluding remarks.
Chapter Preview
Top

Fast Analytics

In this section we address the speed challenge. When the speed of results is a factor to consider for efficient analytics, and taking into consideration that it must be used as much data as possible for the abovementioned reasons, then one must reduce or eliminate processing bottlenecks. They can be the IO latency and CPU usage. CPU is good at computing mathematical operations and multicore CPU’s enhance their performance. IO depends on the type of disk and RAM. Solid state disks are faster than hard disk drives, but random access memory is faster. A typical hard drive has a latency of 5 ms while RAM has only 100 ns, it is 50000 times faster. But also the faster the more expensive. There is one more rapid type of memory: cache memory, which is a small and expensive memory within the CPU die. Main memory or In-memory refers to the processing of data performed in RAM.

Complete Chapter List

Search this Book:
Reset