Post reply

Processing a partitioned cube

  • December 8, 2004 at 11:56 pm

    #109774

    Using Analysis Services 2000, I have a data cube that is partitioned by year/month and draws against a partitioned view (again partitioned by year/month). The problem I am having is when I process the cube - either partion by partion or the entire cube at once via the process wizard, it seems that Analysis Services is not counting the rows in each partition correctly. If I change the cube and have each partition draw from a view created over each partition's respective fact table as opposed to the partitioned view it still seems to incorrectly count the rows for each partition. Has anyone else encountered anything like this? Is there a way to force a row count by each partition prior to processing?

    Thank you,

    Michael


    mhweiss

  • December 9, 2004 at 12:02 am

    #533132

    I think this is a known limitation.  Basically the row count is never updated - which is quite a big problem if you go about trying to recreate the aggs as the formula/s for doing this take row count into consideration - especially if you build a prototype cube with a single partition using a limited data set, then pump millions of rows into it in production *without* resetting the row count for the agg calcs. I seem to recall one of the mgmt docs (on msdn) suggesting that you use the partition mgr tool to ensure that the counts are accurate/correct.

    Steve.

  • December 9, 2004 at 12:43 am

    #533138

    Thanks Steve. I should have mentioned that I use the partition manager to set the aggregations for each partition. I recall now that I a few days ago I found I was missing a join between two tables in one of my snow flake dimensions. Do you think that could have caused the problem? The weird thing is that when I started processing the cube it would come up with the correct total row count for the partitioned view. Anyway I am reprocessing it now and it seems to be counting the rows correctly. At least for the first two partitions so far.

    Thanks,

    Michael


    mhweiss

Viewing 3 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic. Login to reply