Comments on: Viewing Partition Slice Information

By: Pete Griffiths

Pete Griffiths — Tue, 18 Oct 2011 13:22:41 +0000

Thanks Chris,

Suggestion logged here:
https://connect.microsoft.com/SQLServer/feedback/details/695504/correct-internal-dataid-order-of-dimension-members#tabs

Pete

By: Chris Webb

Chris Webb — Tue, 18 Oct 2011 12:35:22 +0000

Hi Pete,

To make a suggestion like this you need to go to https://connect.microsoft.com/ . Once you’ve done that, post the link back here and other people can vote.

Chris

By: Pete Griffiths

Pete Griffiths — Tue, 18 Oct 2011 11:28:18 +0000

I’ve just been suffering with this issue, Thanks for the procedure, it has been very useful!

I’ve resolved some of my issues by creating some indexes on the source table for the columns used in the select statement issued by SSAS when processing the dimension attributes. This meant that my attributes where in the correct order so the dataids are now predictable.

In doing this I thought that a good solution to this whole issue is for the Analysis Services team to make a small change to the dimension processing routine to include an order by statement in the select distinct used. The order by would be for the field used to sort the attribute members whether it be at the key, name, attributekey or attribute name. These values are all included in the column list for the select (for obvious reasons so it should be a simple change for quite a big benefit.

I don’t know how to suggest this change to microsoft but I thought you might possibly have a bit more gravitas if you also think it is a good idea.

Any comments welcome!

By: Leandro

Leandro — Mon, 03 Nov 2008 13:51:54 +0000

Hi Chris

Thanks !. It is exactly what I was looking for. Your improvement could be made by adding Dataids to an array and then executing below mdx to retrieve unique names and keys, just once record loop is finished. I\’ve put three sample DataIds:

With

Member Measures.MemberKey as [Date].[Calendar Year].currentmember.Properties("Key")
Member Measures.MemberName as [Date].[Calendar Year].currentmember.Properties("Name")
SELECT
{Measures.MemberKey, Measures.MemberName} ON 0,
FILTER([Date].[Calendar Year].members, DataId([Date].[Calendar Year].currentmember) = 2 OR
DataId([Date].[Calendar Year].currentmember) = 3 OR
DataId([Date].[Calendar Year].currentmember) = 4 ) on 1
FROM [Adventure Works]

By: Jesse

Jesse — Thu, 22 Nov 2007 07:14:15 +0000

This is an awesome post. Very, very good stuff.

I too have been fighting slice issues recently and am curious to give this a shot.

Thanks again.

By: Unknown

Unknown — Mon, 12 Nov 2007 13:51:03 +0000

Thanks for this great article on partition slice information. I\’ve been relying on examining the INF*.XML files, which is a much more painful way to go about this process. I\’m going to imliment your ideas, and see if I can\’t expand on them a little. I\’ll post backi any significant improvements I find.

On your closing note on the "partition bug". I raised a ticket with Microsoft some months ago, and have been engaged in a regular battle ever since. i provided them with a detailed explanation, a businesss justification and suggestions. It seems that they just can\’t get it. I\’ve been leading with them to connect me with the Query engine and the Storage engine developers, so that I can explain the benefits to someone who understands. (Last email went out last week.)

Basically, all they need to do is to allow you to sort attributes as they are processed by the storage engine, so that the ordinal come in sorted. Microsoft has suggested that I manually load the dimension information (i.e. via SSIS packages), but the effort involved and the number of cubes I\’d then have to manually manage make this unworkable.

I did manage to create the right indexes in SQL Server to trick the SQL optimizer into sending the rows back roughly sorted. It\’s much better (I have 100 partitions and the extraneous partition scans have dropped 75%), but this of course isn\’t very robust or stable. It\’s very much subject to change.

Why this is such a battle, when the benefits are quite obvious I can\’t understand. MSAS is a wonderful piece of technology, but Microsoft is going to have to fix their support processes.

I\’d be happy to provide you with the specifics, if it\’ll help get Microsoft implement this smal change.