[ https://issues.apache.org/jira/browse/FLINK-178?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Fabian Hueske closed FLINK-178. ------------------------------- Resolution: Duplicate Is subsumed by issues around FLINK-986 to cache intermediate results. > Memory-Temping Data Sink / Source > --------------------------------- > > Key: FLINK-178 > URL: https://issues.apache.org/jira/browse/FLINK-178 > Project: Flink > Issue Type: Improvement > Reporter: GitHub Import > Labels: github-import > Fix For: pre-apache > > > One of Spark's most popular features is caching the result of computations in > memory. This enables much more interactive data analysis. > We could add a similar feature by proving a memory-temping data sink and data > source. The Pact runtime already offers a ResettableReader which writes data > into memory and eventually to disk. Currently, all data structures are > destroyed as soon as a job finishes. Keeping such a ResettableReader, would > enable to read data from memory that was computed from a previous job. > However, there are also some challenges with this approach: > - How much memory should be assigned to temping data sinks > - How is working memory and temping memory split up > - What is done if we run out of memory (eviction strategies) > This feature would benefit a lot from dynamic memory assignment / trading. > ---------------- Imported from GitHub ---------------- > Url: https://github.com/stratosphere/stratosphere/issues/178 > Created by: [fhueske|https://github.com/fhueske] > Labels: core, enhancement, > Created at: Fri Oct 18 15:39:39 CEST 2013 > State: open -- This message was sent by Atlassian JIRA (v6.3.4#6332)