Why such an overhead for system with docker containers usage? - docker

Question

I have a following question. I was designing recently a java application on Spring, that works with a database. And I have decided to perform a stress testing. Both the application and the database reside on a virtual Debian machine. I tested it with gatling, and here is what I got:
request count 600 (OK=600 KO=0 )
min response time 12 (OK=12 KO=- )
max response time 159 (OK=159 KO=- )
mean response time 21 (OK=21 KO=- )
std deviation 13 (OK=13 KO=- )
response time 50th percentile 17 (OK=17 KO=- )
response time 75th percentile 22 (OK=22 KO=- )
mean requests/sec 10.01 (OK=10.01 KO=- )
t < 800 ms 600 (100%)
800 ms < t < 5000 ms 0 ( 0%)
t > 5000 ms 0 ( 0%)
failed 0 ( 0%)
So far, so good. Ater that, I decided to put the database and jar into two containers. Here is a docker-compose.yml sample for that:
prototype-db:
build: prototype-db
volumes:
- ./prototype-db/data:/var/lib/mysql:rw
- ./prototype-db/scripts:/docker-entrypoint-initdb.d:ro
ports:
- "3306"
prototype:
image: openjdk:8
command: bash -c "cd /deploy && java -jar application.jar"
volumes:
- ./application/target:/deploy
depends_on:
- prototype-db
ports:
- "8080:8080"
dns:
- 172.16.10.1
- 172.16.10.2
The Dockerfile looks like this:
FROM mysql:5.7.15
ENV MYSQL_DATABASE=document \
MYSQL_ROOT_PASSWORD=root \
MYSQL_USER=testuser \
MYSQL_PASSWORD=12345
EXPOSE 3306
Now, after testing that with gatling I've got the followin results:
---- Global Information --------------------------------------------------------
request count 6000 (OK=3946 KO=2054 )
min response time 0 (OK=124 KO=0 )
max response time 18336 (OK=18336 KO=77 )
mean response time 5021 (OK=7630 KO=10 )
std deviation 4136 (OK=2478 KO=9 )
response time 50th percentile 6516 (OK=8694 KO=9 )
response time 75th percentile 8732 (OK=8905 KO=14 )
mean requests/sec 87.433 (OK=57.502 KO=29.931)
---- Response Time Distribution ------------------------------------------------
t < 800 ms 65 ( 1%)
800 ms < t < 5000 ms 532 ( 9%)
t > 5000 ms 3349 ( 56%)
failed 2054 ( 34%)
---- Errors --------------------------------------------------------------------
java.io.IOException: Remotely closed 1494 (72.74%)
status.find.is(200), but actually found 500 560 (27.26%)
This is amazing - the mean response time exceeded drastically, and a lot of errors, but this docker compose system runs on the very same virtual debian machine. What could cause exactly such an overhead, I thought that docker containers are a lot like native processed, they should not be running that slow.