Migration from SQLServer In Database Machine Learning Services to Standalone RServer RRS feed

  • Question

  • I'm currently running a configuration training and running machine learning models from SQLServer (2016) RServices (task queues using Service Broker, SSIS for data import, and a few stored procedures to prepare data and actually run models) . The host database is used a lot for other processes, and due to the huge ram/cpu requirements of machine learning, I am considereing setting up an additional either SQLServer Standalone RServer or SQL2017 instance for Rservices.

    I'm however unfamiliar with the standalone server, and from what I read in the documentation, it's primary use is (correct me if I'm wrong) to deport computing "nearest-to-the-data" .

    So is there even a point to use a standalone server in this configuration ? Or should I have an SQL2017 instance on my new server and stick with indatabase RServices ? (Or both ? why?)

    In the first case, I don't find in the documentation any examples of stored procedures pushing data to / calling R computations from a standalone server.  (I'm guessing changing context in a sp_executeExternalScript ?)

    Hope to get some advices and explanations of both elements best use-case to have the most efficient setting.

    Thanks in advance,

    Thursday, June 28, 2018 2:02 PM

All replies

  • Hi Chalmel,

    I would advice you to continue using SQL Server R/ML Services (in-database) in another instance of SQL Server. We do not recommend using Standalone R/ML Server when your data is in SQL Server since Standalone R/ML Server does not have the benefits that accrue from tight integration with SQL engine. You already have everything working with in-database R Services so it will be straight-forward for you to replicate this on another instance and you will continue to benefit from the in-database features (like using the T-SQL stored procedures to execute your models).


    Standalone R/ML Server is useful for scenarios when your data is somewhere else (non-SQL Server data sources). The Standalone R/ML Server included in SQL Server setup is just another way of installing Machine Learning Server and is functionally equivalent (e.g. scalable ML and operationalization of the models using web-services etc.) to the matching version of Machine Learning Server.



    Sumit Kumar

    Monday, July 9, 2018 11:37 PM