linux - Eliminating the impact of UnionFS on results when benchmarking inside Docker -
i trying benchmark overall system performance of running docker using phoronix test suite 6.4.0 milestone 2 running inside fedora:23 image based container.
one thing must considered is, docker uses proprietary unionfs store data. however, when running real-world application (like apache) inside docker, persistent data stored on dedicated folder on host, running on standard linux filesystem ext4, or in case btrfs.
the solution propose use "docker volume" mount host directory docker. thing don't know directories used in benchmarks , have mounted inside docker container.
the test suite pts/disk example should use docker volumes instead of unionfs. contains these tests.
pts/compress-gzip pts/sqlite pts/apache pts/pgbench pts/compilebench pts/iozone pts/dbench pts/fs-mark pts/fio pts/tiobench pts/postmark pts/aio-stress pts/unpack-linux
which directories in docker container should mounted host (made docker volumes)? idea use docker volumes? there other caveats consider when benchmarking docker?
as mentioned in "mount shared-storage volume data volume"
in addition mounting host directory in container, docker volume plugins allow provision , mount shared storage, such iscsi, nfs, or fc.
since docker 1.9, done docker volume create
(and mount volumes in expected paths in container)
that allows to:
- make volumes independent of host
- still bypass ufs
- experiments different docker volume plugins
- compare their performance benchmarks
which directories in docker container should mounted host
any folder include persistent data, and/or has high volume of io operations.
Comments
Post a Comment