Uploaded image for project: 'Marathon'
  1. Marathon
  2. MARATHON-3528

performance issues with docker images

    Details

    • Type: Task
    • Status: Resolved
    • Priority: Medium
    • Resolution: Won't Do
    • Affects Version/s: None
    • Fix Version/s: None
    • Component/s: None
    • Labels:

      Description

      I noticed this strange behavior for a docker image which performs some cpu intensive tasks:

      If I start the docker container via marathon with a cpus-setting of 0.1 this seems to clamp the cpu utilization for that container to roughly 10% max even if no other stuff is running on that node. Setting cpus to 0.1 results in setting cpu-shares to 204 for the docker run command (verified with docker inspect).

      However, if I start the exact same docker container manually with docker run --cpu-shares=204 ... on the same node the container can utilize the cpu by ~100% (as expected).

      I don't understand why both containers (spawn from the exact same image) are behaving differently with respect to cpu utilization. I couldn't find any significant differences in the docker inspect outputs.

      According to the various docs the behavior of the pure docker run meets my expectations and understanding: the cpus/cpu-shares settings are just hints about the ratio for distributing cpu cycles in case of cpu-overutilization; if there is only one running task asking for cpu cycles it should get the full cpu power without clamping/throttling, right?

        Attachments

          Activity

            People

            • Assignee:
              Unassigned
              Reporter:
              GitHub_Kosta-Github Kosta (Inactive)
              Team:
              Orchestration Team
              Watchers:
              Jason Gilanfarr (Inactive)
            • Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: