Mark Rittman has worked up an excellent article on the new Oracle BI Landscape. He basically covers the whole range of products and technologies and how they’re being repackaged and rebranded into a new product set. Great read.
I thought the following was of interest:
The Analytic Server is effectively a ROLAP (Relational OLAP) engine, that connects to data sources but does not have any local storage in the traditional, database sense, although it does have a cache which holds copies of data previously retrieved, so that subsequent requests for the same data are returned faster.
What struck me about this statement is that from a technology perspective this is nearly identical to Mondrian‘s (Open Source OLAP engine) architecture. While proprietary vendors (Oracle included) had been touting the benefits of a true MOLAP solution instead of just ROLAP. However, with the processing headed to mid tier servers (instead of OLAP crunching happening on a Pentium I desktop) one can leverage the mid tier for caching, and analytical processing (MOLAP-esque stuff). In other words, you can get MOST of the benefits of MOLAP with a well designed and performant ROLAP server proactively caching/crunching on your behalf! Cool stuff!
Hi Nick.
Good observation. One key difference though is that Mondrian has (I think) an MDX interface, whilst OracleBI Analytics Server has a relational, ODBC interface to the outside world.
Another thing to bear in mind is that, from my impressions anyway, the Analytics Server isn’t seen as a particularly “good” OLAP server, certainly not as fast/functional/capable as a true MOLAP such as MS AS or Oracle OLAP. In most cases, Analytics Server will pass OLAP queries back to the back-end OLAP server to execute on their own MOLAP datasets, it only has very rudimentary OLAP functionality within the mid-tier although Oracle are looking to supplement this going into the future.
So there is a difference there between Analytics Server and Mondrian – mondrian has it’s own OLAP engine within the mid-tier which it uses all the time, whilst Oracle’s new offering will try and offload the OLAP processing to the source OLAP databases when at all possible, meaning that there’s still a role for traditional OLAP servers in their architecture. I don’t think they see the mid-tier taking on the role of an OLAP server, it’s more of an integration layer and cache for the applications.
cheers
Mark
Mark — Thanks for the comments! Seeing as I’m not as “in touch” with the Oracle BI product teams your articles/comments help keep me informed! 🙂
That’s really interesting what you say… I’m in agreement that MOLAP will offer superb performance in terms of slicing and dicing analytic cubes (time series, versus period prior, percentage as total of top N products, etc etc). With the availability of copious amounts of RAM I wonder how a ROLAP server would fare… but I don’t know! As I connect with more advanced users of Mondrian perhaps I can report back how they find it. I’m sure there are copious numbers of academic papers as well proving that ROLAP can never be as fast as MOLAP in certain dimensional analysis’.
Now I’m trying to figure this out… It is inherently TOUGH to express multidimensional data sets and queries using SQL. Oracle has used the OLAP_TABLE (is that right, I forget) table function which was akward at best (just my opinion). Given that the anlaytics server can speak “MDX” on the back end how is this expressed on the front end using ODBC/SQL? If they’ve found a way to do this (express multi-dim in SQL) eloquently that’s a huge win for customers then! 🙂
btw, thanks for helping me understand how it’s all coming together… the more products/techs that come together (with significant overlap) tough to keep who does what straight!
Cheers!
> Now I’m trying to figure this out… It is inherently TOUGH to express multidimensional data sets and queries using SQL. Oracle has used the OLAP_TABLE (is that right, I forget) table function which was akward at best (just my opinion). Given that the anlaytics server can speak “MDX” on the back end how is this expressed on the front end using ODBC/SQL? If they’ve found a way to do this (express multi-dim in SQL) eloquently that’s a huge win for customers then!
Well, what the Siebel server does is convert the multidimensional datasets from MS AS etc into relational rowsets, and then just process and join them as any other set of relational data. This presentation here:
http://db.cs.berkeley.edu/dblunch-fa2005/siebel.ppt
(“Modeling and Querying Multidimensional Data Sources in Siebel Analytics”)
describes the process quite well. My impression is that Siebel (or nQuire before them) got lots of plaudits for doing this, in that you can bring multi-dimensional data into what is essentially a relational system, but what you get as the end product is a relational, row-based query environment. You’ve lost much of the rich OLAP metadata, or at least the Siebel query tools don’t give you much of an OLAP-style environment to work with. My understanding is that rectifying this is the #1 priority for Oracle, to add some multidimensional “smarts” to the platform by bringing in some of the OLAP functionality of Oracle Discoverer.
cheers
Mark