UNICOM Intelligence Reporter Server Edition provides different performance levels than the Desktop version, because it relies on shared server resources and overall network status. When opening a document in UNICOM Intelligence Reporter Server Edition, the entire document (metadata and tables) must be transferred to the client. Similarly when populating tables, the table definitions are transferred to the server and the results back to the client. Because the table generation process is on more powerful server, it is possible that UNICOM Intelligence Reporter Server Edition can populate the tables faster than the Desktop version.

The objective of this white paper is to summarize the UNICOM Intelligence Reporter Server Edition limits, that are based on a series of test results, and to provide guidelines for effective UNICOM Intelligence Reporter Server Edition usage.

Test 1: Open MDD files that work with minimal case data

All variables in the MDD files are categorical, and each variable has 20 categories.

The results show that when the variable count of the MDD increases to more than 5000, the time to open UNICOM Intelligence Reporter Server Edition increases to approximately 60 seconds (1 minute).

The client memory usage increases fairly linearly. The memory usage is acceptable and it is likely that users will encounter problems with the open time before they encounter memory usage problems.

The following tables are used:

Because all variables in the test metadata contain 20 categories, the following table sizes are the result:

Because the case data size is very small, the data aggregation process does not consume much time. For the 160,000 cell table, approximately 22 seconds is used to display the table in the Results tab. The reason is that UNICOM Intelligence Reporter Server Edition is reliant on Microsoft Internet Explorer. No significant improvement is seen when running the 160,000 cell table on UNICOM Intelligence Reporter Desktop Edition.

The first time populating any table in UNICOM Intelligence Reporter Server Edition requires additional time due to the case data component initialization on the server. The test compares the initial population with subsequent populations against a simple table (20 * 20 (400 cells)).

The results show that the time to initially populate a table is considerable. The behavior is completely independent of the actual table that was populated. In cases where the MDD variable count can increase to more than 2000, it is advisable to use a small table when first starting UNICOM Intelligence Reporter Server Edition.

Using an MDD file that includes 2000 multiple-response variables, and 23,350 records in the case data (DDF format), serial tables were created with incremental element counts (for example, 20*20, (10+10)*20, (20+10)*20...).

When populating tables without statistical tests, UNICOM Intelligence Reporter Server Editionrequires the same amount of time to generate results for each kind of table structure. The maximum amount of time required is 2 - 3 seconds. However, the table structure significantly impacts overall performance when populating tables that include column proportions statistics.

The following chart shows the time usage (in seconds) for specified elements that are contained on the side and top axis. For example, the first cell has a value of 4.88, which means that table 20*20 needs 4.88 seconds to populate. The last cell shows that the table (20+20+20+20)* (20+20+20+20) takes 273 seconds to populate.

The test results provide the following conclusions:

The test includes 24 similarly structured tables that are defined in UNICOM Intelligence Reporter Server Edition. No statistical tests are applied to the tables (but the tables do include filters).

The following table presents the time required to populate each table:

The results show that case data that is stored in PKD files are, on average, 5% slower than case data that is stored in DDF files. DDF is the optimum case data choice for UNICOM Intelligence Reporter Server Edition.

The tests were run on two UNICOM Intelligence server versions:

The results show that UNICOM Intelligence 7.0.1 is, on average, 10% faster than UNICOM Intelligence 6.0.1 when using DDF case data. Although every table has the similar structure 2437*17 (41429 cells), the population time usage values are not as close. This is possibly due to different filters that are defined for the various tables. Two elements become apparent when examining each table filter: variable count and condition count. The variable count was tested and found not to be a factor (the document omits the result). When considering the condition count, all boolean expressions are counted in the filter. For example, the following filter syntax is treated as 6 conditions.

The condition counts that are defined for each filter are as follows:

The following chart shows the relationship between the table filter conditions and the table population time in the UNICOM Intelligence 7.0.1 environment:

The data fits a fairly linear trend. Using more filter conditions results in more table population time. Care must be taken when increasing the condition count in the table or global filters.

The MDD file includes 2000 multiple-response variables. 7 MTD documents, that each contain 800 tables, were created. For each table, the structure has incremental element counts (for example, 20*20, 20*20+20, 20+20*20+20...). The table cells only vary between the different variables, and all of tables contain no stats or filters. The test records two data attributes: opening the MTD, and then populating the tables.

1 Get warn: The maximum request length was exceeded, because the populate request string is too large to invoke remote UNICOM Intelligence Reporter Server Edition with the default Web Service buffer.

The results show that loading 800 tables requires less than 30 seconds, even when the table structure grows very large. The table population performance decreases significantly when more variables are introduced onto an axis. The default Web service request limits prevent the successful population of 800 tables when the table structure is larger than (1600 cells, 4 variables).

11 MTD documents are created to contain an incremental number of tables (800, 1600, 2400, 3200...), but all table structures are 20*20 (400 cells). Similar to Test 7, the tables attempt to use various metadata variables as much as possible, and all tables contain no filters. The test also records two data attributes: opening the MTD, and then populating the tables.

All tables are 20*20 and include stats (grid overlap)

1 The accurate time should be more than 12 hours. The server process w3wp.exe hung for several hours after the memory usage reached 2GB. All tables populated without error.

Populating all tables is extremely slow after the UNICOM Intelligence Reporter Server Edition table number exceeds 800. As a result, only 800 and 1600 tables have a recorded population time. The test server was limited to 2GB of memory. More than 8GB memory would allow the 1600 tables to be populated within an hour.