Add readrandom with both memtable and sst regression test

Summary: @MarkCallaghan's tests indicate that performance with 8k rows in memtable is much worse than empty memtable. I wanted to add a regression tests that measures this effect, so we could optimize it. However, current config shows 634461 QPS on my devbox. Mark, any idea why this is so much faster than your measurements?

Test Plan: Ran the regression test.

Reviewers: MarkCallaghan, dhruba, haobo

Reviewed By: MarkCallaghan

CC: leveldb, MarkCallaghan

Differential Revision: https://reviews.facebook.net/D14511
This commit is contained in:
Igor Canadi 2013-12-11 13:51:20 -08:00
parent a8029fdc75
commit f5f5c645a8
1 changed files with 99 additions and 17 deletions

View File

@ -26,15 +26,20 @@ function cleanup {
}
trap cleanup EXIT
git_br=$(basename $GIT_BRANCH)
if [ -z $GIT_BRANCH ]; then
git_br=`git rev-parse --abbrev-ref HEAD`
else
git_br=$(basename $GIT_BRANCH)
fi
if [ $git_br == "master" ]; then
git_br=""
else
git_br="."$git_br
fi
make clean
OPT=-DNDEBUG make db_bench -j$(nproc)
make release
# measure fillseq + fill up the DB for overwrite benchmark
./db_bench \
@ -45,7 +50,8 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--num=$NUM \
--writes=$NUM \
--cache_size=6442450944 \
--cache_numshardbits=6 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--statistics=1 \
--histogram=1 \
@ -60,9 +66,10 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--use_existing_db=1 \
--bloom_bits=10 \
--num=$NUM \
--writes=$((NUM / 2)) \
--writes=$((NUM / 10)) \
--cache_size=6442450944 \
--cache_numshardbits=6 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--statistics=1 \
--histogram=1 \
@ -80,7 +87,8 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--num=$NUM \
--writes=$NUM \
--cache_size=6442450944 \
--cache_numshardbits=6 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--statistics=1 \
--histogram=1 \
@ -96,9 +104,10 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--use_existing_db=1 \
--bloom_bits=10 \
--num=$NUM \
--reads=$NUM \
--reads=$((NUM / 5)) \
--cache_size=6442450944 \
--cache_numshardbits=8 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--disable_seek_compaction=1 \
--statistics=1 \
@ -106,18 +115,19 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--disable_data_sync=1 \
--disable_wal=1 \
--sync=0 \
--threads=32 > ${STAT_FILE}.readrandom
--threads=16 > ${STAT_FILE}.readrandom
# measure readrandom with 300MB block cache
# measure readrandom with 100MB block cache
./db_bench \
--benchmarks=readrandom \
--db=$DATA_DIR \
--use_existing_db=1 \
--bloom_bits=10 \
--num=$NUM \
--reads=$NUM \
--cache_size=314572800 \
--cache_numshardbits=8 \
--reads=$((NUM / 5)) \
--cache_size=104857600 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--disable_seek_compaction=1 \
--statistics=1 \
@ -125,7 +135,70 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--disable_data_sync=1 \
--disable_wal=1 \
--sync=0 \
--threads=32 > ${STAT_FILE}.readrandomsmallblockcache
--threads=16 > ${STAT_FILE}.readrandomsmallblockcache
# measure readrandom with 8k data in memtable
./db_bench \
--benchmarks=overwrite,readrandom \
--db=$DATA_DIR \
--use_existing_db=1 \
--bloom_bits=10 \
--num=$NUM \
--reads=$((NUM / 5)) \
--writes=512 \
--cache_size=6442450944 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--write_buffer_size=1000000000 \
--open_files=55000 \
--disable_seek_compaction=1 \
--statistics=1 \
--histogram=1 \
--disable_data_sync=1 \
--disable_wal=1 \
--sync=0 \
--threads=16 > ${STAT_FILE}.readrandom_mem_sst
# fill up the db for readrandom benchmark with filluniquerandom (1GB total size)
./db_bench \
--benchmarks=filluniquerandom \
--db=$DATA_DIR \
--use_existing_db=0 \
--bloom_bits=10 \
--num=$((NUM / 4)) \
--writes=$((NUM / 4)) \
--cache_size=6442450944 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--statistics=1 \
--histogram=1 \
--disable_data_sync=1 \
--disable_wal=1 \
--sync=0 \
--threads=1 > /dev/null
# measure readrandom after load with filluniquerandom with 6GB block cache
./db_bench \
--benchmarks=readrandom \
--db=$DATA_DIR \
--use_existing_db=1 \
--bloom_bits=10 \
--num=$((NUM / 4)) \
--reads=$((NUM / 4)) \
--cache_size=6442450944 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--open_files=55000 \
--disable_seek_compaction=1 \
--disable_auto_compactions=1 \
--statistics=1 \
--histogram=1 \
--disable_data_sync=1 \
--disable_wal=1 \
--sync=0 \
--threads=16 > ${STAT_FILE}.readrandom_filluniquerandom
# measure memtable performance -- none of the data gets flushed to disk
./db_bench \
@ -135,7 +208,8 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--num=$((NUM / 10)) \
--reads=$NUM \
--cache_size=6442450944 \
--cache_numshardbits=8 \
--cache_numshardbits=4 \
--table_cache_numshardbits=4 \
--write_buffer_size=1000000000 \
--open_files=55000 \
--disable_seek_compaction=1 \
@ -145,13 +219,19 @@ OPT=-DNDEBUG make db_bench -j$(nproc)
--disable_wal=1 \
--sync=0 \
--value_size=10 \
--threads=32 > ${STAT_FILE}.memtablefillreadrandom
--threads=16 > ${STAT_FILE}.memtablefillreadrandom
# send data to ods
function send_to_ods {
key="$1"
value="$2"
if [ -z $JENKINS_HOME ]; then
# running on devbox, just print out the values
echo $1 $2
return
fi
if [ -z "$value" ];then
echo >&2 "ERROR: Key $key doesn't have a value."
return
@ -180,5 +260,7 @@ send_benchmark_to_ods overwrite overwrite $STAT_FILE.overwrite
send_benchmark_to_ods fillseq fillseq $STAT_FILE.fillseq
send_benchmark_to_ods readrandom readrandom $STAT_FILE.readrandom
send_benchmark_to_ods readrandom readrandom_smallblockcache $STAT_FILE.readrandomsmallblockcache
send_benchmark_to_ods readrandom readrandom_memtable_sst $STAT_FILE.readrandom_mem_sst
send_benchmark_to_ods readrandom readrandom_fillunique_random $STAT_FILE.readrandom_filluniquerandom
send_benchmark_to_ods fillrandom memtablefillrandom $STAT_FILE.memtablefillreadrandom
send_benchmark_to_ods readrandom memtablereadrandom $STAT_FILE.memtablefillreadrandom