FranckPachot / DB-Engines-CPU

A tool to measure the CPU usage when running the same workload on different databases (this is not a benchmark)

DB-Engines-CPU

A tool to measure the CPU usage when running the same workload on different databases

⚠️ this is not a benchmark:

• what is run on each may not be equal (especially the Sakila scripts are different for each DB)
• we account for the background processes during the run so a longer response time counts more of them
• the configuration of the databases are not tuned equally (please suggest improvement)

Usage

The Docker Compose defines how to start the database engines and their CLI.

For example, you can start YugabyteDB and create the Sakila schema:

docker compose down
docker compose up yugabytedb -d --wait
docker compose run -T yugabytedb-cli < ./sakila/yugabytedb-sakila-db/yugabytedb-sakila-schema.sql

To insert Sakila data and get the number of CPU instructions with Linux Perf Stat, we identify the CGROUP and call perf while running the script:

 container=$(docker-compose ps -q yugabytedb )
 ls -ld /sys/fs/cgroup/perf_event/docker/$container
  perf stat -e instructions -G docker/$container -a \
   docker compose run -T yugabytedb-cli < ./sakila/yugabytedb-sakila-db/yugabytedb-sakila-insert-data.sql

It displays the number of CPU instructions counted by perf stat -e instructions for the whole engine (group):

Run all scripts on all databases

You can run:

sh run-all.sh

and check the elapsed time and CPU instructions

(
cd out
awk '
BEGIN{e["postgres"]="🐘";e["oracle"]="🅾️ ";e["yugabytedb"]="▝▞";e["cockroachdb"]="🪳";}
{ split(FILENAME,f,"/") ; gsub(f[1],"*",f[2]) }
/^ +[0-9,]+ +instructions +docker[/][0-9a-f]+/{ ins=$1 }
/^ +[0-9.]+ +seconds time elapsed/ { printf "%-25s %30s instructions in %6.1fs %15s %1s\n",f[2],ins,$1,f[1],e[f[1]] }
' */small* | sort  -k1,1 -k2,2n | awk '$1>l{print ""}{print}{l=$1}'
) | tee out/summary.txt