|
|
| What is a
Metric? |
Metrics are also called "facts" or "measures." These are the quantitative components of a multi-dimensional structure. In a relational model these are numeric non-keyed columns. Any numeric column that can be aggregated can be a metric.
QueryObject supports several different kinds of aggregation, such as simple sum, max, min, filtered etc.
Any column can become a distinct count metric, which tells about the number of distinct items found in the column for each multidimensional coordinate.
QueryObjects can have a maximum of 100 metrics. |
|
|
| Is there a limitation on the number of dimensions and metrics that can be handled by a QueryObject? |
| A single QueryObject can contain up to 15 true dimensions, any number of hierarchies attached, up to 100
metrics. |
|
|
| Is there a theoretic limit also for the QO Ready File? |
| 2^32-1
dimensions, metrics and keyback columns |
|
|
| Is
there a limitation on the number of Values /
Categories in a Dimension? |
| 2^32-1
categories. |
|
|
| What is
Tresholding? |
|
Query Objects are made up of a number of n-dimensional InfoStores, where n ranges from 1 to the number of dimensions in the QueryObject.
For example, if the dimensions are A, B and C, the QueryObject contains the following sub-InfoStores:
| InfoStore
No. |
Composition |
Level |
| 1 |
ABC (a.k.a Unique Level Keys) |
3 |
| 2 |
AB |
2 |
| 3 |
AC |
2 |
| 4 |
BC |
2 |
| 5 |
A |
1 |
| 6 |
B |
1 |
| 7 |
C |
1 |
|
If we specify that the engine is to perform thresholding, certain levels of InfoStores will not appear in the QueryObject. This means that thresholding removes sub-InfoStores in groups having the same number of dimensions.
For example, if we threshold the above using Low=2, and High=3, we get the following QueryObject:
|
InfoStore
No. |
Composition |
Level |
| 1 |
ABC (a.k.a Unique Level Keys) |
3 |
| 2 |
AB |
2 |
| 3 |
AC |
2 |
| 4 |
BC |
2 |
From experience (and theory) we know that the actual size of the levels is illustrated by the drawing.
The x- axis represents the number of the level (1 to 15), the y axis represents its relative size in the QueryObject. Obviously its best to threshold low, leaving out the middle levels (in a 1-15 type QueryObject, the biggest level would be around 8)
However if local KeyBack support is needed, the system will automatically include the highest level, which is needed by the KeyBack process.
Support for Thresholding
Thresholding works on levels, not particular dimensions, so if you have a QueryObject with thresholding set from 1 to 3, all you have to do is keep in mind not to use more then 3 dimensions in one query. You can still use up to any 3 of the 15 in each query.
Of course if you set thresholding at 2 to 3 for example, you will need to provide at least 2 dimensions in a query (or drill down sequence in our Viewer tool) to receive any results.
|
|
|
| Is it possible to predict the dimension of a QO before launching the Build
process? |
| Not precisely. Since we store only information not data, the size of the files depends on how varied your input data is, what kind of dimensions you use, how many (and which) dimensions, keybacks and metrics you design into your
QO. |
|
|
| How long does it take to build a QO? How long does it take to build a QO
ODS? |
In many accounts QO technology was used because it was the only one able to completely load data and create infostores in the allotted time window. The exact performance depends on many factors, such as disk speed, CPU speed, number of CPUs (scales very well on SMP machines), amount of memory.
The performance of the system is under constant improvement so each new release of the system works faster then the previous one. Often these are very significant performance
gains. |
|
|
|
Is it possible to manage hierarchies in a dynamic way? |
Yes, hierarchies can be created and attached to an existing
cube.
Multiple hierarchies can be attached to each base
dimension.
Hierarchies have no limit on the number of levels, other then the combinatorial limit of 2^64-1 on the product of the number of members across all levels attached to a single base
dimension. |
|
|
|
Will the software evolve towards data access methods different from
ODBC? |
We already support ODBO access, though do not recommend it. Once the Java based OLAP API standard matures, we may implement a provider.
We already support access using a very simple and convenient C API , and a Java version of that API is also available.
We have developed our own C/.NET API to support developing front end clients that use a specific view paradigm. This is used in our QO Reader product that is bundled with the QO Personal and Workgroup
edition. |
|
|
