Skip to content

don't check current key in DBIter::Next() - #679

Merged
pwnall merged 1 commit into
google:masterfrom
smartxworks:optimize-readseq
May 4, 2019
Merged

pwnall merged 1 commit into
google:masterfrom
smartxworks:optimize-readseq

Conversation

@kylezh

@kylezh kylezh commented Apr 25, 2019

Copy link
Copy Markdown
Contributor

When iter_ is pointing to current key, we can safely move to the next
key to avoid checking current key, which is of course not necessary.

Benchmark shows that 'readseq' has about 8% performance improvement.

Without patch:

./db_bench --benchmarks=readseq --num=$((4<<20)) --db=/tmp/db --use_existing_db=1
LevelDB: version 1.21
Date: Thu Apr 25 09:37:21 2019
CPU: 32 * Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
CPUCache: 20480 KB
Keys: 16 bytes each
Values: 100 bytes each (50 bytes after compression)
Entries: 4194304
RawSize: 464.0 MB (estimated)
FileSize: 264.0 MB (estimated)


readseq : 0.196 micros/op; 565.7 MB/s

With patch:

./db_bench --benchmarks=readseq --num=$((4<<20)) --db=/tmp/db --use_existing_db=1
LevelDB: version 1.21
Date: Thu Apr 25 09:38:20 2019
CPU: 32 * Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
CPUCache: 20480 KB
Keys: 16 bytes each
Values: 100 bytes each (50 bytes after compression)
Entries: 4194304
RawSize: 464.0 MB (estimated)
FileSize: 264.0 MB (estimated)


readseq : 0.181 micros/op; 612.3 MB/s

Signed-off-by: Kyle Zhang kyle@smartx.com

When iter_ is pointing to current key, we can safely move to the next
key to avoid checking current key, which is of course not necessary.

Benchmark shows that 'readseq' has about 8% performance improvement.

Without patch:

>./db_bench --benchmarks=readseq --num=$((4<<20)) --db=/tmp/db --use_existing_db=1
LevelDB:    version 1.21
Date:       Thu Apr 25 09:37:21 2019
CPU:        32 * Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
CPUCache:   20480 KB
Keys:       16 bytes each
Values:     100 bytes each (50 bytes after compression)
Entries:    4194304
RawSize:    464.0 MB (estimated)
FileSize:   264.0 MB (estimated)
------------------------------------------------
readseq      :       0.196 micros/op;  565.7 MB/s

With patch:

>./db_bench --benchmarks=readseq --num=$((4<<20)) --db=/tmp/db --use_existing_db=1
LevelDB:    version 1.21
Date:       Thu Apr 25 09:38:20 2019
CPU:        32 * Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
CPUCache:   20480 KB
Keys:       16 bytes each
Values:     100 bytes each (50 bytes after compression)
Entries:    4194304
RawSize:    464.0 MB (estimated)
FileSize:   264.0 MB (estimated)
------------------------------------------------
readseq      :       0.181 micros/op;  612.3 MB/s

Signed-off-by: Kyle Zhang <kyle@smartx.com>
@ghemawat

Copy link
Copy Markdown

Thanks. I looked carefully and this looks correct to me.

@kylezh

kylezh commented May 4, 2019

Copy link
Copy Markdown
Contributor Author

@cmumford Could you take some time to look at this patch?

@pwnall
pwnall merged commit d3d1c8a into google:master May 4, 2019
pwnall added a commit that referenced this pull request May 4, 2019
@pwnall

pwnall commented May 5, 2019

Copy link
Copy Markdown
Contributor

@kylezh Thank you for your contribution!

@cmumford brought this change in the internal repository. I just ran the commands that brought github back in sync.

maochongxin pushed a commit to maochongxin/leveldb that referenced this pull request Jul 21, 2022
The windows SDK headers don't have self-consistent casing anyway,
and many projects consistently use lowercase for them, in order
to fix crosscompilation with mingw headers.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants