You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kudu.apache.org by "Todd Lipcon (Code Review)" <ge...@ip-10-146-233-104.ec2.internal> on 2016/01/21 00:31:43 UTC
[kudu-CR] cbtree: some optimizations to make back ground since gcc 4.4
Hello Adar Dembo,
I'd like you to do a code review. Please visit
http://gerrit.cloudera.org:8080/1848
to review the following change.
Change subject: cbtree: some optimizations to make back ground since gcc 4.4
......................................................................
cbtree: some optimizations to make back ground since gcc 4.4
Benchmarks indicated that cbtree-test got slower when we upgraded to gcc
4.9:
4.4 (pre-C++11):
I0120 14:04:21.391079 1982 cbtree-test.cc:730] Time spent Insert 4000000 keys: real 2.934s user 2.868s sys 0.064s
I0120 14:04:25.186092 1982 cbtree-test.cc:730] Time spent Insert 4000000 keys: real 2.856s user 2.852s sys 0.004s
I0120 14:04:28.971114 1982 cbtree-test.cc:730] Time spent Insert 4000000 keys: real 2.852s user 2.856s sys 0.000s
I0120 14:04:32.786329 1982 cbtree-test.cc:730] Time spent Insert 4000000 keys: real 2.851s user 2.856s sys 0.000s
I0120 14:04:36.601415 1982 cbtree-test.cc:730] Time spent Insert 4000000 keys: real 2.850s user 2.852s sys 0.000s
4.9.3 (C++11):
I0120 14:31:10.460209 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.266s user 3.212s sys 0.028s
I0120 14:31:14.625602 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.191s user 3.168s sys 0.000s
I0120 14:31:18.799840 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.199s user 3.176s sys 0.004s
I0120 14:31:23.010079 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.198s user 3.180s sys 0.000s
I0120 14:31:27.220559 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.198s user 3.172s sys 0.004s
I0120 14:31:31.434557 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.203s user 3.184s sys 0.000s
I0120 14:31:35.648149 9809 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.201s user 3.176s sys 0.004s
Looking at the new assembly code, I found a few easy improvements to win back
some of the performance. With this patch:
I0120 14:31:59.956502 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.188s user 3.120s sys 0.044s
I0120 14:32:04.061563 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.125s user 3.104s sys 0.000s
I0120 14:32:08.173352 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.136s user 3.104s sys 0.012s
I0120 14:32:12.290326 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.141s user 3.116s sys 0.004s
I0120 14:32:16.396093 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.131s user 3.104s sys 0.004s
I0120 14:32:20.510488 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.141s user 3.120s sys 0.000s
I0120 14:32:24.769479 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.278s user 3.256s sys 0.000s
I0120 14:32:28.866739 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.125s user 3.104s sys 0.000s
I0120 14:32:33.185539 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.346s user 3.320s sys 0.000s
I0120 14:32:37.284078 9851 cbtree-test.cc:732] Time spent Insert 4000000 keys: real 3.124s user 3.120s sys 0.000s
which is a few percent gain, but still missing a good chunk.
Change-Id: I63afea81ce9d4ab605ccb27d1d9345637613c06d
---
M src/kudu/tablet/concurrent_btree.h
M src/kudu/util/inline_slice.h
2 files changed, 4 insertions(+), 4 deletions(-)
git pull ssh://gerrit.cloudera.org:29418/kudu refs/changes/48/1848/1
--
To view, visit http://gerrit.cloudera.org:8080/1848
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-MessageType: newchange
Gerrit-Change-Id: I63afea81ce9d4ab605ccb27d1d9345637613c06d
Gerrit-PatchSet: 1
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Todd Lipcon <to...@apache.org>
Gerrit-Reviewer: Adar Dembo <ad...@cloudera.com>