Have you guy consider using Apache Arrow format (http://arrow.apache.org/) as the in-memory tuple batch structure for vectorized execution? I think the goal of Apache Arrow project matches the one of vectorized execution perfectly, and its community is quite active, although the implementation of Apache Arrow format is far away from the implementation of PostgreSQL/GPDB/HAWQ tuple structure , which means the engineering efforts needed would be huge.
Just my two cents. Lirong Lirong Jian HashData Inc. 2018-02-26 13:40 GMT+08:00 Hongxu Ma (JIRA) <[email protected]>: > Hongxu Ma created HAWQ-1591: > ------------------------------- > > Summary: Common tuple batch structure for vectorized execution > Key: HAWQ-1591 > URL: https://issues.apache.org/jira/browse/HAWQ-1591 > Project: Apache HAWQ > Issue Type: Sub-task > Components: Query Execution > Reporter: Hongxu Ma > Assignee: Lei Chang > Fix For: backlog > > > A common tuple batch structure for vectorized execution, holds the tuples > which be transfered between vectorized operators. > > > > > > -- > This message was sent by Atlassian JIRA > (v7.6.3#76005) >
