Hello Andrew Wong,
I'd like you to do a code review. Please visit
http://gerrit.cloudera.org:8080/8554
to review the following change.
Change subject: wire_protocol: optimize RewriteRowBlockPointers
......................................................................
wire_protocol: optimize RewriteRowBlockPointers
I was looking at why our tpch1 benchmark performance seems to have been
creeping up over recent months and started profiling it a bit. I didn't
see any reason for a regression but did notice that
RewriteRowBlockPointers ends up being memory-bandwidth-bound, so took a
few minutes to optimize it.
The old code did one pass through the data for each column. In the case
that the rowblock was larger than CPU cache, this meant streaming a lot
of data from L2, L3, or even memory, which bottlenecked it. The new code
is branchier but does only a single pass through the data.
This reduced the client side CPU usage of the query in tpch_real_world
from about 70ms/iteration to about 40ms/iteration.
Change-Id: Ie428e595e9f564762bbdbd09dc6a5f312abe9aec
---
M src/kudu/common/wire_protocol.cc
1 file changed, 48 insertions(+), 29 deletions(-)
git pull ssh://gerrit.cloudera.org:29418/kudu refs/changes/54/8554/1
--
To view, visit http://gerrit.cloudera.org:8080/8554
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: newchange
Gerrit-Change-Id: Ie428e595e9f564762bbdbd09dc6a5f312abe9aec
Gerrit-Change-Number: 8554
Gerrit-PatchSet: 1
Gerrit-Owner: Todd Lipcon <t...@apache.org>
Gerrit-Reviewer: Andrew Wong <andrew.w...@cloudera.com>
