Chris Webb's BI Blog

Analysis Services, MDX, PowerPivot, DAX and anything BI-related

Archive for the ‘Random Thoughts’ Category

Enterprise Search and BI

with one comment

I notice from various sources (for example, Don Dodge) that Microsoft have released a free version of their Enterprise Search product, Microsoft Search Server 2008 Express. The thing that caught my eye was the list of federated connectors:

…which includes Business Objects, Cognos, SAS, but there’s no mention of Reporting Services or Analysis Services anywhere. As I think I’ve said here before, I’m not convinced that a search interface on top of a BI platform is going to be useful in the real world (though I bet you could do some cool demos with it): I suppose if you have hundreds of SSRS reports for example you might want to look for the ones that contain figures for a particular Product or Customer, but I would have thought that it’s just as likely that you’d do a search, find a report and then find you don’t have permission to view it. As for using a search interface as a way of querying a cube, all I have to say about that is two words: English Query.

But I think there’s a more interesting application for BI here: what if you could build a cube off the index this thing creates? You could have dimensions like Date Updated, Keyword, File Type and Path, and measures like Count of Files and File Size; you’d be able to do things like create reports which tracked the overall space taken by mp3 files on your network and where these files were, the number of emails with the phrase "new job" in; even just browsing ad hoc in Excel you’d have a new way of searching for files: for example, you could slice on File Type=Word doc, Keyword="CV" or "Resume" and then put the Path dimension on rows and drill down to find all the CVs on your network.

Written by Chris Webb

November 8, 2007 at 10:06 pm

Posted in Random Thoughts

Adapting SQLIS to work with tuples and sets as a data source

with 3 comments

It’s been a long time since I posted in my ‘random thoughts’ category… but I just had such an interesting idea I thought I’d post it up (even if there’s 0% chance I’ll ever get round to implementing this).
I was looking at a (non-Microsoft platform) BI tool today and got thinking about MDX, how people find it hard to work with, and how most client tools don’t really expose the power of MDX sets, and how handy it would be to be able to do some procedural things in MDX too. This particular tool had some cool set-based selection functionality and I reflected that even though I’d seen similar set-based selection tools, some on AS (didn’t Proclarity have something in this area?), they’d never really taken off; I also thought about the much-missed MDX Builder tool which had a similarly visual approach to building MDX expressions. I started thinking about whether it would be worth building another client tool which took this approach but quickly came to the conclusion that the world needed another AS client tool like a hole in the head, but realised that if I was going to build this kind of tool how much it would resemble Integration Services. And then I had my idea: why not extend Integration Services so it can treat MDX sets and tuples as a data source, and then use its existing functionality and create new transformations to implement MDX set-based operations?
Let me explain in more detail. I’m not talking about simply getting data out of AS in the same way you’d get it out of a SQL Server table, using an MDX query. What I’m saying is that what would be flowing though the IS data flow tasks would be members, sets and tuples: each ‘row’ of data would be an MDX expression returning member, or tuple, or set. So you’d create a custom data source where you could define a set as your starting point – probably at this point you’d just select a whole level, or the children of a member, or some such simple set of members. For example you might select the [Customers].[Customer].[Customer] level in your Customer dimension; the output from this would be a single text column and a single row containing the set expression [Customers].[Customers].[Customers].Members. You could then put this through an Exists() transform to return only the customers in the UK and France, the output from which would be the set expression Exists([Customer].[Customer].[Customer].Members, {[Customer].[Country].&[United Kingdom], [Customer].[Country].&[France]}). Similarly then you could put this through a Crossjoin() transform to crossjoin this set with the set of all your Products, then put the result through a NonEmpty() transform to remove all non empty combinations from the set. At this point your output would still be a single row and column, consisting of the MDX expression:

, {[Customer].[Country].&[United Kingdom], [Customer].[Country].&[France]})
, [Product].[Product].[Product].Members)
, [Measures].[Internet Sales Amount])

So far, so dull though. All we’ve got is a way of building up a string containing an MDX set expression and SQLIS brings little to the party. But the real fun would start with two more custom transformations: SetToFlow and FlowToSet. The former would take an input containing MDX set expressions (and conceivably there could be more than one row, although we’ve only got one so far) and would output a flow containing all the tuples in the set(s) we’ve passed in. Taking the set above, the output would be the contents of measures.outputdemo in the following query on AdventureWorks:

with member measures.outputdemo as TupleToStr(
([Customer].[Customer].Currentmember, [Product].[Product].Currentmember)
select {measures.outputdemo} on 0,
, {[Customer].[Country].&[United Kingdom], [Customer].[Country].&[France]})
, [Product].[Product].[Product].Members)
, [Measures].[Internet Sales Amount])
on 1
[Adventure Works]

The FlowToSet transform would do the opposite, ie take an input containing tuples and return a single row containing the set represented by the entire input. For the above example, this would be a big set:
{([Customer].[Customer].&[12650],[Product].[Product].&[214]), ([Customer].[Customer].&[12650],[Product].[Product].&[225]),…}
But the point of this would be that you could then apply more MDX set expressions efficiently, although of course there’s no reason why you can’t apply MDX set expressions to individual tuples in a data flow. The final important
custom transform you’d need would be an Evaluate transform, which would append one or more numeric or text columns to a tuple or set dataflow: each of these columns would be populated by evaluating an MDX expression which returned a value against the set or tuple for each row. So, for example, if a row contained a the set we’ve been using we could apply a the Count function to it and get the value 12301 back; if a row contained the tuple ([Customer].[Customer].&[12650],[Product].[Product].&[214]) we could ask for the value of this tuple for the measure [Internet Freight Cost] and get the value 0.87 back; or to the same tuple we could ask for the value of [Customer].[Customer].CurrentMember.Name and get back the value "Aaron L. Wright".
Of course the beauty of this is that once you’ve got a flow containing sets, tuples and numeric values retrieved from the cube for them then you can use all the cool existing SQLIS functionality too, like multicasts, lookups, UnionAlls, Aggregates etc to do stuff with your sets that is hard in pure MDX; and of course you can easily integrate other forms of data such as relational or XML, and do useful things at the end of it all like send an email to all your male customers in the UK who bought three or more products in the last year, or who live in London and have incomes in excess of £50000 and have averaged over £50 per purchase, or who have been identified as good customers by a data mining model, and who aren’t on the list of bad debtors that you’ve got from the Accounts department’s Excel spreadsheet.
Now of course all of this is possible with using only relational data with SQLIS, or even without using SQLIS and just using pure MDX. I guess the point of this is, as always, that it provides an easier way to do stuff: build MDX expressions without having to know much MDX, integrate AS data with other data and other applications without doing (much) coding, and so on.
So, as ever, I’d be interested in your comments on this. I have the distinct feeling that this is a solution in search of a problem… but if you can think of some problems it might solve, then let me know!

Written by Chris Webb

August 25, 2006 at 3:52 pm

Posted in Random Thoughts

Microsoft, BI and Search

with one comment

It’s inevitable, when you get a whole bunch of new functionality as we have with SQL2005, that you start thinking of the new types of applications that become possible. One of the things I’ve been thinking about for a while is how you could take the results of an RSS feed or a search engine search, do text mining on the results and build a cube to analyse what comes back. Quite an interesting idea, I think, and I know plenty of other people have been thinking along the same lines too, eg
And it’s not just in the Microsoft world that these ideas are cropping up. For example, only today I saw a reference to a (non-Microsoft) OLAP solution which built cubes from the results of text mining:
Anyway, on a different note, one of the fun things about blogging is all the rumours and snippets of information about new solutions coming soon, most of which I’m not really at liberty to discuss (not that I know much anyway). You get to put these snippets, rumours and other stuff you read on the web and put them together in a 1+1=3 operation… Here, for example, is a link that Jon-who-sits-next-to-me just sent which he saw on Slashdot:
How can Microsoft beat Google in the search game? There are some interesting hints on the second page of this article, for example:

He said that Microsoft’s goal — but not its initial offering — would go beyond finding URLs and instead focus in on the specific information sought by Internet users.

"Generally these days what you get back is URLs, and based upon research 50 percent of the time you do a search you don’t get the URL you’re looking for," he said.

Holloway said that the promise of Microsoft’s search capability is to dig down.

For example, he said, potential home-buyers might find a group of houses in the price range and with the precise amenities they are seeking.

Or a surfer might find a restaurant with the kind of menu a diner wants in a particular geographic area.

Hmm, is it me or is there a potential BI angle here? Dig down == drill down, perhaps? Slice, dice and analyse your resultset rather than just get a flat list of links? I wonder… 
UPDATE: Jon, bless his heart, has come up with another interesting link on this topic:
Don’t you just love wild speculation? The whole Origami thing is so last week… 
UPDATE#2: Now this could just be me reading way too much into something, but here’s another relevant link:
There’s a coincidence here that’s too good to be true…

Written by Chris Webb

March 2, 2006 at 2:24 pm

Posted in Random Thoughts

Reward Beta Testers better!

with 2 comments

Before I even write this I know I’m going to sound ungrateful, whingeing, grasping, greedy and all sorts of other things, but here goes…
There was a SQL2005 launch event in London yesterday, and one of my colleagues from another team went along. He sent the link to register to the event to me and some others on my team, but we had a look at the agenda and we realised that we’d seen pretty much all of the material before so we didn’t go. I started working with AS2005 in the Spring of last year, and when I joined my current team my colleagues had already been working with AS2005 for more than a year before; as a result we know quite a lot about AS2005 and as beta testers we found a lot of bugs. Anyway, I just spoke to the guy who went to the launch event this morning and he told me that he got a FREE copy of SQL2005 Standard Edition and a FREE copy of Visual Studio Professional just for turning up. And I admit that I was jealous. Yes I got a fleece when I went to my first Yukon airlift back in 2002(?), yes I got a Yukon t-shirt and picture frame last year, and yes I got a *lot* of inside knowledge, help and support as a beta tester which has been invaluable professionally and which I’m very, very grateful for, but copies of SQL2005 and VS are really quite tasty gifts. You’d think that Microsoft would at least treat beta testers as well as the people that go to launch events by giving them a free copy of the product they’ve been testing, wouldn’t you? 

Written by Chris Webb

November 16, 2005 at 10:58 am

Posted in Random Thoughts

Usage-Based Partitioning

leave a comment »

I was reading Dave Wickert’s excellent white paper "Project REAL: Analysis Services Technical Drilldown" the other day (you can get it here), specifically the section on P39 about partitioning. In it he discusses the new functionality in AS2005 which automatically determines which members from your dimensions have data in a given partition, and goes on to talk about the new possibilities this opens up in terms of partitioning strategy. Here’s an excerpt:

The partitions in the Project REAL database seem to violate one of the basic best practices of SQL Server 2000. There is no data slice set for the partitions. In SQL Server 2000, partitions must have the data slice set so that the run-time engine knows which partition to access. This is similar to specifying a hint to a relational query optimizer. In SQL Server 2005, this is no longer necessary. Processing the partition now automatically builds a histogram-like structure in the MOLAP storage. This structure identifies which members from all dimensions are included in the partition. Thus, so long as the storage method is MOLAP, the data slice is an optional (and unused) property. However, the data slice is used with ROLAP storage or when proactive caching involves a ROLAP access phase. In both of these circumstances, the actual fact data is never moved so the system does not have a chance to identify a member. In this case, setting the data slice for the partition remains a necessary and critical step if you expect the system to perform well.

Because the MOLAP structures dynamically determine the data slice, a new type of partitioning technique is possible with SQL Server 2005. The best way to describe this technique is via a simple example.

Suppose a system that you are designing has a product dimension of 1,000 products. Of these, the top 5 products account for 80% of the sales (roughly evenly distributed). The remaining 995 products account for the other 20% of the sales. An analysis of the end-user query patterns show that analysis based on product is a common and effective partitioning scheme. For example, most of the reports include a breakdown by product. Based on this analysis, you create six partitions. You create one partition each for the top 5 products and then one “catchall” partition for the remainder. It is easy to create a catchall partition. In the query binding, add a WHERE clause to the SQL statement as in the following code.

In the top five partitions (1 through 5) use the following code.

      SELECT * FROM <fact table>
      WHERE SK_Product_ID = <SK_TopNthProduct#>

In the catchall partition use the following code.

      SELECT * FROM <fact table>
      WHERE SK_Product_ID NOT IN (<SK_TopProduct#>,




This technique requires a lot of administrative overhead in SQL Server 2000 Analysis Services. In SQL Server 2000, the data slice must identify each and every member in the partition—even if there are thousands and thousands of members. To implement the example, you would need to create the catchall partition data slice with 995 members in it. This is in addition to the administrative challenge of updating that list as new members are added to the dimension. In SQL Server 2005 Analysis Services, the automatic building of the data slice in the partition eliminates the administrative overhead.

 This got me thinking… if we’ve got a Usage-Based Optimisation wizard for helping design the right aggregations for a cube, surely it’s possible to do something similar so that we can design partitions on the basis of the queries that users actually run? Here’s an idea on how it might work (nb this would be a strategy to use in addition to partitioning by Time, Store or other ‘obvious’ slices rather than a replacement):
  • First, get a log of all the queries that users are actually running. Unfortunately the Query Log in AS2005, like AS2000, doesn’t actually record the actual MDX of all the queries run. The only way to do this is to use Profiler; I was a bit worried about whether doing this would have an adverse impact on query performance but when I put the question to MS they indicated it shouldn’t be much (Richard Tkachuk also mentioned, as an aside, that turning off Flight Recorder should result in an increase of a few % in terms of query performance – a tip to remember for production boxes, I think). Once you’ve run your trace you can then export all of the MDX statements from it to a text file very easily.
  • You’d then need a bit of code to extract the unique names of all the members mentioned explicitly in these queries. It should be a fairly simple task if you get the right regular expression, I think. Note that this ignores queries which use any kind of set expression – my thinking was that individually named members are going to be the most interesting because they’re going to be the ones which slice the queries the most: if users are querying on all the countries in the cube that’s not going to be any use for partitioning, but if they have a particular product in the WHERE clause that is much more useful to know about.
  • Then you could do some data mining to cluster these members by their propensity to appear in a query together. The idea is that each of the resulting clusters would translate into a partition; those members which didn’t fall nicely into a cluster and those members that didn’t get extracted in step #2 would have their data fall into one of Dave’s ‘catch-all’ partitions. Imagine this scenario: the UK branch of the Adventure Works corporation suddenly finds there is massive demand for bikes after petrol (‘gas’, for you Americans) prices rise massively. As a result, analysts in the UK run lots of queries which are sliced by the Product Category [Bikes] and the Country [UK]. You’d hope that this pattern would emerge in the clustering and result in a single partition containing all the data for ([Bikes], [UK]), so in the future similar queries run much faster.

What does everyone think? There seems to be a lot of activity these days in the comments section of my blog, so I thought I’d invite feedback. Can anyone see a fatal flaw in this approach?


Written by Chris Webb

October 20, 2005 at 2:54 pm

Posted in Random Thoughts

LinkShare: my idea for a $50000-prize winning app

leave a comment »

As I said, I don’t have the time to enter the Connected Systems Developer competition that I blogged about the other week, but that hasn’t stopped me thinking about what I might build if I did enter. The following idea came to me at around 2am this morning when I was desperately trying to get my 22-month-old daughter to go back to sleep, and having nothing better to do this afternoon I thought I’d bounce it off anyone reading my blog. So comments are invited – even if they are just to say that it’s a rubbish idea and/or someone’s thought of it before and/or it’ll never work. I have after all categorised this post under ‘Random Thoughts’!

Business Case:

In the modern office everyone does a lot of web surfing; some of it might even be business-related. And whenever we see something interesting we typically copy the link into a mail, add a few words of explanation and send it on to a few people who might also want to have a look. I send at least two or three such emails a day. For the typically lazy web surfer, though, this process is a bit of a hassle so we only bother to do it when we think the link is really interesting and (because we don’t want to get a reputation as the office spammer) we only send it to a small number of people we know who we think are going to find it interesting too. It’s my contention that it would be cool if we could share more of these links with more people.

So, we need to solve three problems in our quest to share the interesting links we find during our daily surfing:
1) It needs to be easier to share the link once we’ve found it
2) We don’t want to send stuff on to people who aren’t going to be interested, and we want to receive only the links that we’re going to be interested in.
3) We’d like to be able to share links with people who we don’t actually know well enough to contact directly.

Of course there are plenty of existing ways that people share links, such as newsgroups, email discussion lists and blogs but they typically only address the third of the above problems fully, the second only partially and the first not very well at all. For instance, anyone reading my blog is presumably doing so because they’re interested in Microsoft’s BI tools and they’re going to be interested in any links to webcasts, articles etc that I post up, but if they’re like me they subscribe to upwards of a hundred rss feeds – and that’s only on subjects they’re really interested in – so we still have the proverbial information overload. The same goes for email discussion lists and newsgroups. And in all these cases in order to share information you have to open an email, write a blog post etc, which all require effort.

Let me give you an idea of the kind of scenario I want to tackle. This morning I was reading this story on the Register, and followed a link on a whim to this page, a set of pictures of Cybermen with funny captions. It brought a smile to my face but I didn’t send it on to anyone else because a) it didn’t seem worth the bother, and b) I didn’t know whether any of the people I usually send stuff onto were at least mildly into Dr Who in the way I am. I’m not going to blog about it because it’s not relevant to BI, I don’t subscribe to any Dr Who blogs, dls or newsgroups because I’m not that much of a Dr Who fan, and so no-one else is going to see it. Which is a shame.

Functional Spec:

Anyway, enough waffle about the theory. The solution I’m thinking of would consist of something like the following:

  • An IE toolbar with only two controls on it: a button saying "This is a cool page" which you hit whenever you find an interesting link (regardless of whether you’ve found it yourself or it has been recommended to you by the system), and a textbox which allows you to add a short commentary on the contents of the page if you want. Whenever you hit the button, it sends the current url in your browser plus any comments to a web service which…
  • …Puts the information in a queue on a server. There’s an app which gradually works its way through every link submitted, retrieves the page, strips the text from the html and does some funky text mining on this and the comments you’ve submitted, and classifies it. This is then used by…
  • …Another server app which looks at your tastes (based on pages you’ve submitted in the past and perhaps other users who you’ve said you’re interested in seeing links from) and then using some more data mining gives you a short list of recently submitted links that you might be interested in, along with the comments of the people who have recommended them. This could either come in the form of a web page, a customised rss feed or a regular email newsletter.

So, in practice, let’s imagine it working as follows. Chris, Jon and Colin all work in a large corporation, in the same team doing the same kind of BI stuff. During his morning surfing, Chris submits 5-10 links; one, on a new feature of MDX, gets recommended automatically to Jon and Colin because everyone in the team works with MDX and has submitted MDX-related pages in the past. One, containing pictures of Cybermen with amusing comments, gets recommended only to Colin and only appears about halfway down his list because he’s a bit of a sci-fi fan and has submitted a few sci-fi links in the past. Meanwhile, David, who works in a different team and doesn’t know Chris, Jon or Colin finds a cool article on C-Omega and submits it so it gets recommended to the rest of his team; they all in turn click their buttons and so it eventually appears at the top of Jon’s list (because he’s really into coding) and somewhere down the list for Chris (because he’s not so into coding, but this is a really cool article nonetheless).

The larger the number of users with similar taste, the better it should work – more links submitted plus more people voting on the same links, and so the mining models can get to know people’s tastes much more quickly. I could imagine it doing well as an intranet app at a large tech company. It would probably need to give more priority to newer links (people want the latest stuff, and you don’t want old but popular links clogging up your recommendations) and maybe have some way of removing links you’ve already seen from your list of recommendations. One other extra feature that occurred to me was that the app could also generate a report showing the users who submitted the most interesting links, so as to generate a bit of rivalry and encourage future usage.

The key to it all though is the fact that all you need to do to submit a page is click a button in IE – the absolute minimum effort possible – and the fact that the job of the mining model is clear – recommend a page which will make you click your button in turn.


It should be fairly straightforward to build the toolbar and the web service. Qualification for the competition comes with the use of SQL 2005 for storing all the data, SQLIS to do the processing, AS to do the data mining, and RS to do the web-based reports, daily email, even the rss feed (maybe as a custom rendering extension?). I’ll admit that I don’t know enough about data mining to know whether that bit will really work, but hey, it might.


OK, enough fantasising. If anyone does implement this and enters the competition, please can I have a share of the winnings?


Written by Chris Webb

June 17, 2005 at 4:22 pm

Posted in Random Thoughts

Grouping in MDX – response to Mosha

with one comment

I’m honoured by my mention in Mosha’s blog! But I think my entry on Grouping in MDX, and Mosha’s comments on it, need some further clarification and discussion.

First of all, CREATE SESSION CUBE. I agree it is probably the best way to implement grouping at the moment and that it works well in Excel, but it’s not ideal:

  • Surely it’s a client-only solution? You’d want the option of creating these groupings on the server for all users, just as you would any other calculated member, and CREATE SESSION CUBE doesn’t offer you that, does it?
  • I don’t know what the overhead of creating a session cube on the client is (or, in Yukon, on the server) – but is it as small as creating a calculated member? What is the impact on caching, especially given the fact that you have to drop and create the session cube when you change or add a new group?
  • A lot of the time, for example in Reporting Services, you are only working with one-off queries rather than sessions and you really want to be able to create a group with query rather than session scope.
  • The MDX needed to create the session cube and then manage it within the session seems overly complex for such a simple requirement.

Secondly, to address Mosha’s point on why you would need the VisualTotals and Aggregate functions in the same query: it’s because you’d want to use your new group member in any scenario where you could use a normal member, and that includes a query which used VisualTotals. Imagine you had a measure which showed the distinct count of customers across all your stores, and you wanted a report which had a) a single group member containing your top 5 stores, b) several other individual stores and c) showed the visualtotal of all the distinct customers in both the group and the individual stores displayed. I think that would be a reasonable requirement and one which wouldn’t be possible unless AS ‘knew’ what members went into the group.

Thirdly, sets in the WHERE clause (and also subcubes in the FROM clause, which do the same thing) inYukon. Unfortunately, this only works when you’re slicing by the group and not when the group is on a visible axis so it doesn’t fit the scenario I was describing.

Overall, then, CREATE SESSION CUBE is almost the functionality that I want but it doesn’t allow groups to be defined on the server. So we’re close…!

Written by Chris Webb

January 24, 2005 at 2:47 pm

Posted in Random Thoughts

Grouping members together

with 2 comments

One of the weaknesses of Analysis Services, in my opinion, is support for creating custom groupings of members. I reckon that 90% of all calculated members on non-measures dimensions must be doing just this, ie just doing an AGGREGATE or SUM over a set of members, and yes calculated members will return the right values but my complaint is something else. It’s that you then have no idea what members were aggregated together inside this calculated member, and that functions like VISUALTOTALS, NONEMPTYCROSSJOIN etc that you would like to be ‘group aware’ of course aren’t. Some examples needed, I think…

Consider the following query on Foodmart 2000:

WITH MEMBER [Customers].[All Customers].[USA].DEMO AS ‘AGGREGATE({[Customers].[All Customers].[USA].[CA], [Customers].[All Customers].[USA].[OR]})’
{[Customers].[All Customers].[USA].CHILDREN, [Customers].[All Customers].[USA].DEMO} ON 1

Wouldn’t it be nice, then if VISUALTOTALS ‘knew’ what was in the set and this query

WITH MEMBER [Customers].[All Customers].[USA].DEMO AS ‘AGGREGATE({[Customers].[All Customers].[USA].[CA], [Customers].[All Customers].[USA].[OR]})’
VISUALTOTALS({[Customers].[All Customers].[USA],{[Customers].[All Customers].[USA].[CA], [Customers].[All Customers].[USA].[OR]} },  "* DEMO")
ON 1

returned the same results as this query?

WITH MEMBER [Customers].[All Customers].[USA].DEMO AS ‘AGGREGATE({[Customers].[All Customers].[USA].[CA], [Customers].[All Customers].[USA].[OR]})’
VISUALTOTALS({[Customers].[All Customers].[USA],[Customers].[All Customers].[USA].DEMO },  "* DEMO")
ON 1

And that when you did a NONEMPTYCROSSJOIN against your calculated member, it would return the same results as when you did a NONEMPTYCROSSJOIN against the set that was aggregated in the calculated member? And perhaps also that you could drill down from the calculated member to see the members inside it?

Of course this isn’t possible at the moment, because a calculated member could contain any sort of calculation, so AS simply can’t make any assumptions. But if there was a special kind of group calculated member, which simply took a set of members as its definition and which always returned an AGGREGATE of that set, surely AS could make these assumptions?

Just a thought…

Written by Chris Webb

January 12, 2005 at 4:00 pm

Posted in Random Thoughts


Get every new post delivered to your Inbox.

Join 3,113 other followers