[KUDU-637] Multi-thread KuduScanNode

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Public beta
• Component/s: impala, perf
• Target Version/s: Public beta
• Assignee: Unassigned
• Reporter: Todd Lipcon
• Created: 05/Mar/15 12:20
• Updated: 03/Sep/15 19:02
• Resolved: 03/Sep/15 19:02

Description

Currently the KuduScanNode uses blocking APIs to read data from Kudu. This means that, for many queries, although we use the same number of CPU cycles as Parquet, we spend a lot more wall clock time, which makes us look unimpressive.

We should switch to the simplest multi-threaded model that gets our performance in range of Parquet.