Github user StephanEwen commented on the pull request: https://github.com/apache/flink/pull/1003#issuecomment-131403660 This is again baked into the Flink runtime. Is there a way to keep this separate? I am still a bit puzzled if this is the pattern with which people use parameter servers: embedded in the workers that operate on the training data. My impression so far was that the parameter servers are a separate component, usually running on different machines. Also, could you comment on how this could work with the changes proposes in #967
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---