Introducing Microsoft's OLAP Server
There are OLAP servers (OLAP software that runs primarily on the server) and OLAP clients (OLAP software that runs primarily on the client). There are ROLAP, multidimensional OLAP (MOLAP), and hybrid OLAP servers. As Table 2 shows, Microsoft’s new OLAP server sits in the middle.
Figure 1 shows the architecture of this OLAP server architecture. (Figure 1 is reproduced with permission from The OLAP Report, published by Business Intelligence, at www.olapreport.com.) Here is a description of its parts:
A
: Applications can be written in any language that can consume object linking and embedding (OLE) automation, including Visual Basic, VB Script, Java, Java Script, Delphi, PowerBuilder, C, and C++. However, OLAP-aware objects are not likely to be available in the short term.
B: The next version of Microsoft Office for Windows will provide basic OLAP client facilities. Microsoft will use extended pivot tables that can access the data cube service (DCS) via OLE DB for OLAP standard to achieve these client facilities.
C: Numerous third party clients and applications will support the OLE DB for OLAP standard.
D: Non-OLAP-aware SQL query and reporting tools will be able to view cubes as flattened, fully denormalized SQL tables.
E: All access to the OLAP server is via the DCS. A proprietary protocol that distributes both data and metadata to local caches links the DCS to the OLAP server. The DCS can run on the server or the client. The DCS performs all multidimensional calculations other than aggregations.
F: The DCS can access any OLE DB sources directly for small-scale desktop OLAP applications.
G: An OLAP application can have multiple, independent physical cubes, each with its own dimensions. Dimensions can be shared between cubes or private.
H: Each cube can (optionally) have multiple data partitions, possibly sourced from different databases. Partitions can use ROLAP, MOLAP, or hybrid OLAP architectures and individually optimized or cloned sparse aggregation strategies. The OLAP server aggregates data from partitions on-the-fly.
I: Virtual cubes are like relational views: They store no data and are formed by joining two or more physical cubes on the common dimensions (if any). Virtual cubes can include additional calculated members of their own.
J: The server can source data and dimensions directly from any database accessible via OLE DB (and thus ODBC). In practice, this means just about every relational database running on any platform, plus some OLAP servers, including those from Microsoft, Informix, Seagate Software, and Oracle.
K: In ROLAP partitions, SQL automatically creates the sparse aggregates, which are stored as tables in the source relational database. In hybrid OLAP and MOLAP partitions, the OLAP server automatically creates the sparse aggregates, which are stored multidimensionally. Regardless of the storage and aggregation strategy, the functionality is unchanged.
Anticipating the OLAP Server's Effects
According to The OLAP Report’s in-depth January 15, 1998, analysis, "OLAP Server: Microsoft Corporation," Microsoft's OLAP server will change the OLAP marketplace. The report states: "It is no surprise that the Microsoft OLAP server can be expected to transform the OLAP industry. It will be distributed on a mammoth scale, at prices that are a small fraction of those charged for current OLAP servers. It will have all the ease-of-use features that are expected in any modern Microsoft product. It will have unrivaled client tool support and will probably be enthusiastically adopted by legions of Microsoft solution providers and specialist application builders. What is probably much more surprising to many is that this is a product that is good enough to have been a winner even without the power assistance from Microsoft's formidable marketing machine. In practical terms, it is the first OLAP solution that can take full advantage of the capabilities of all three tiers in a modern client/server environment….There are question marks, of course. There will be some applications which would work much better with a more server-centric calculation approach, particularly if large dimensions are to be analyzed….Despite the question marks, all the signs are that this product will meet the needs of the majority of OLAP users. It will be easier to implement than most of today's large-scale servers, and will integrate well with both Microsoft and other relational databases.
Just how inexpensive will Microsoft's OLAP solution be compared to competitors' solutions? The OLAP Report states that, "Overnight, Microsoft clearly intends to turn OLAP servers into a commodity. The legendary Microsoft business model of generating huge revenues through selling enormous volumes of software products at low prices will certainly be used for the new OLAP Server. Although final prices for SQL Server 7.0 are not known, they are likely to be a small fraction of the prices charged for today's high-end OLAP serversÖproducts like such as Essbase, Holos, DSS Server or Gentia are priced close to $100,000 region for a ten concurrent user server license (including essential, but unbundled, utilities). Lower priced servers, like Express, TM1, MetaCube or Media/MR are priced in the $25,000-$50,000 bracket for a ten concurrent user system. [The new IBM DB2 for OLAP bundles start at $25,000 or $32,000 for the workgroup and enterprise versions, for example, but developers can purchase the personal developer edition for $1,300—KW]. Even if someone bought SQL Server 7.0 purely for its OLAP Server, the prices are likely to be significantly less than a tenth of those currently charged for OLAP servers."
I wholeheartedly agree with The OLAP Report’s assessment. Based on the Beta 2 version I have used, I am extremely impressed with the care Microsoft has put into ease of use. Microsoft claims to have embraced a one-size-doesn't-fit-all approach that lets users select among MOLAP, ROLAP, and hybrid OLAP storage options. The OLAP server has several wizards, including one that helps construct the new OLAP data store, using the DTS engine that makes it easy to select data from any OLE DB or ODBC data source. Another wizard will analyze your proposed model for sparsity to help you make decisions about the number of dimensions and pre-calculations. Microsoft OLAP server is a true multicube OLAP, so you can join a number of cubes to form databases and calculations can involve multiple cubes.
Mark my words: This product will change your life. What’s more, it promises to be incredibly good for a 1.0 release.