[ https://issues.apache.org/jira/browse/PHOENIX-1073?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14056427#comment-14056427 ]
James Taylor commented on PHOENIX-1073: --------------------------------------- Probably best to start on the dev list to discuss this first, rather than file a JIRA. Yes, HBase is much faster if all the data is in it's cache. > A memory table in every region is needed? > ----------------------------------------- > > Key: PHOENIX-1073 > URL: https://issues.apache.org/jira/browse/PHOENIX-1073 > Project: Phoenix > Issue Type: Wish > Reporter: jay wong > > When a do a group by query. > We assume that a Region has 30M data. > 100K row which include 30 kv per row > And the RT of GroupedAggregateRegionObserver is about 3 sec. > but most of time. in fact nearly 2.2 sec is spend on RegionScaner scan all > the row. > I have a test. first time scan all of the data into memory. > the second time only load the data from memory. > the RT of GroupedAggregateRegionObserver execute only 0.7s. > So If a memory table is needed for Phoenix computational intensive scene -- This message was sent by Atlassian JIRA (v6.2#6252)