Best soluton for NoSQL OLAP RRS feed

  • Question

  • Hi I'm trying to figure out what the best solution is for Analytical capabilities on unstructured datasets.

    My ultimate goal is to model relationship data in a way that allows me to define the relationship at run-time rather than hard-code it into the design of the database as with traditional RDBMS. My current thinking is to the store the data in CosmosDB as I won't know the table structure up front (it is likely to be star-schema esque (or facts and dims for those that are familiar with data warehousing), but I won't know the exact column headers up front, nor will every row have every column populated.

    However NoSQL databases are not good for joins + aggregations I realize so am looking into what other options are out there.

    I currently have the following items on my list of technologies to research:

    1. SQL Server 2019 - I believe fields can be of type JSON and therefore flexible, but wasn't sure if joins can be set at run-time

    2. HDInsight Interactive Query (https://docs.microsoft.com/en-us/azure/hdinsight/interactive-query/apache-interactive-query-get-started)

    3. Issuing remote queries to Databricks  Azure Analysis Services

     4. Commercial OLAP offerings available through HDInsight :

    a.       https://kyligence.io/ (fast OLAP for Hadoop – built using Apache Kylin, which is open-source, but the version on HDInsight seems to be the commercial offering)

    b.      https://www.atscale.com/ (another OLAP solution – this one looks pretty slick, but I’m sure it comes at a cost)

    5. SSAS + ComsosDB: However not sure about the dynamic joins

    6. Azure Data Explorer:  https://azure.microsoft.com/en-us/blog/azure-data-explorer-technology-101/

    Any advice on the right architecture here?

    Monday, December 17, 2018 3:59 PM