none
Tabular Model In Server or Client Memory

    Question

  • Hi,

    Could someone clarify whether the tabular model is stored in the server memory or client memory? If it is in the client memory, what happens if the client does not have enough memory -- does the performance suffer terribly?

    Thanks

    MB

    Friday, December 06, 2013 5:23 PM

Answers

  • You don't get memory warnings, only errors :). Also you don't get it unless you process the model (that is just because the server doesn't know how big your model is going to be) or running some complex calculations.

    You might get memory errors during

    - when you process the data in SSDT (development time). Now if you have the same data as your production and if you process that successfully when you are building the model, chances are you will not face memory issues. (provided you don't have complex calculated fields or calculated columns)

    - when you process the data in prod. You might have developed the model with only 10% of the data that prod has. But now when you load across all years, you might run into memory issues. Also, it is recommended to have atleast 3 times the size of the processed model in RAM space (as it needs extra space to process the table and perform calculations).

    - when you run complex calculated fields which might need to iterate through large tables or across multiple tables repetitively.

    It is best to read through this whitepaper - Hardware sizing a tabular model - for sizing your hardware.

    http://msdn.microsoft.com/en-us/library/jj874401.aspx


    Cheers,
    Jason | www.SqlJason.com
    P.S. : Please click the 'Mark as Answer' button if a post solves your problem! :)

    Some Random Thoughts

    Follow me on Twitter

    • Marked as answer by A_B_M Tuesday, December 10, 2013 10:37 AM
    Monday, December 09, 2013 6:48 PM
  • Hi MB,

    Tabular models are in-memory databases in Analysis Services. And it’s on the server. Assume it’s on the client, when multiple users connect the server, it will load the database to the client computer’s memory which will cause the network congestion problem. So t Tabular models are in-memory databases on the server rather than on client. For the detail information about tabular models, please refer to the following link.
    Tabular Modeling (SSAS Tabular)

    Regards,


    Charlie Liao
    TechNet Community Support

    • Marked as answer by A_B_M Tuesday, December 10, 2013 10:37 AM
    Sunday, December 08, 2013 3:23 PM

All replies

  • Hi MB,

    Tabular models are in-memory databases in Analysis Services. And it’s on the server. Assume it’s on the client, when multiple users connect the server, it will load the database to the client computer’s memory which will cause the network congestion problem. So t Tabular models are in-memory databases on the server rather than on client. For the detail information about tabular models, please refer to the following link.
    Tabular Modeling (SSAS Tabular)

    Regards,


    Charlie Liao
    TechNet Community Support

    • Marked as answer by A_B_M Tuesday, December 10, 2013 10:37 AM
    Sunday, December 08, 2013 3:23 PM
  • Thanks for this. And what if the tabular model is too large to fit into the memory on the server? Does SSAS warn the developer during build time if this is the case?
    Monday, December 09, 2013 10:52 AM
  • You don't get memory warnings, only errors :). Also you don't get it unless you process the model (that is just because the server doesn't know how big your model is going to be) or running some complex calculations.

    You might get memory errors during

    - when you process the data in SSDT (development time). Now if you have the same data as your production and if you process that successfully when you are building the model, chances are you will not face memory issues. (provided you don't have complex calculated fields or calculated columns)

    - when you process the data in prod. You might have developed the model with only 10% of the data that prod has. But now when you load across all years, you might run into memory issues. Also, it is recommended to have atleast 3 times the size of the processed model in RAM space (as it needs extra space to process the table and perform calculations).

    - when you run complex calculated fields which might need to iterate through large tables or across multiple tables repetitively.

    It is best to read through this whitepaper - Hardware sizing a tabular model - for sizing your hardware.

    http://msdn.microsoft.com/en-us/library/jj874401.aspx


    Cheers,
    Jason | www.SqlJason.com
    P.S. : Please click the 'Mark as Answer' button if a post solves your problem! :)

    Some Random Thoughts

    Follow me on Twitter

    • Marked as answer by A_B_M Tuesday, December 10, 2013 10:37 AM
    Monday, December 09, 2013 6:48 PM