stress test

[Ibeltek]

Hi all! I have got a quite unexpected performance issue on a LXC guest system. See below stress test results:


i've done test performance on the host system and in the container, result differ 130 times
below is the result of the test on the host system and the container, respectively,
stress-ng —cpu 8 —cpu-ops 80000
stress-ng: info: [6511] successful run completed in 0.30s

stress-ng —cpu 8 —cpu-ops 80000
stress-ng: info: [4034] defaulting to a 86400 second run per stressor
stress-ng: info: [4034] dispatching hogs: 8 cpu
stress-ng: info: [4034] successful run completed in 45.55s

provided below the container configuration:

lxc.arch = amd64
lxc.include = /usr/share/lxc/config/ubuntu.common.conf
lxc.monitor.unshare = 1
lxc.tty = 2
lxc.environment = TERM=linux
lxc.utsname = zakupki-de.com.ua
lxc.cgroup.memory.limit_in_bytes = 68719476736
lxc.cgroup.memory.memsw.limit_in_bytes = 74088185856
lxc.cgroup.cpuset.cpus = 0-7
lxc.cgroup.cpu.shares = 500000
lxc.rootfs = /var/lib/lxc/100/rootfs
lxc.network.type = veth
lxc.network.veth.pair = veth100i0
lxc.network.hwaddr = 00:50:56:00:B1:24
lxc.network.name = eth0

I need to provide a container of 100% server hardware resources

Did someone encounter similar issues? Let me know if there some extra data, which could help to evaluate the issue.

 

[fabian]

note that stress-ng is not a benchmark tool.. are you sure this is not cause by different versions of stress-ng on the host and in the container?

the output at least is different.. stress-ng from ubuntu xenial shows your behaviour, stress-ng from debian jessie does not - so I assume something changed in between