SmartClient Analytics Module: Support for large data sets

[bdandersen]

Hello --

We’re considering introducing analytical capabilities within our SmartGWT application; specifically (in a somewhat contrived fashion), we’d like the user to be able to identify trends in the volume of "events" being generated across time and geographic locations. Naturally, we’re interested in understanding more about the SmartClient Analytics Module.

Have you tested your CubeGrid against data sources that exceed the 50,000 rows of data within your databound_cubegrid example? If so, has the analytics module proven to be reasonably performant when the user (or perhaps multiple users operating in different Web browsers but engaging the same server on the back end) slices frequently (or perhaps rapidly)? What are the module’s realistic limits when it comes to the number of measures in fact tables?

I ask because we anticipate some measures to number in the millions. Continuing the use case cited above, imagine the lowest level of granularity -- the lowest attribute in the time dimension’s hierarchy -- to be seconds; each measure would represent the number of "events" recorded in the course of a second within a specific geographic location.

Any insight into how the analytics module might satisfy our large data set, multiple user scenario would be appreciated.

Thanks and regards,

Brad Andersen
AZAD Consultant - ARRIS Group, Inc.

[Isomorphic]

So long as you make effective use of tree facets such that the user doesnt start with all data already expanded (which isn't particularly useful anyway), you should be fine even with millions of cells.

The earliest limit you are likely hit is when there are around 1000 visible columns in the worst case browser (IE6), at which point you may begin to get script running slowly dialogs. Limits for modern browsers are much higher. You should also prefer to have more rows than more columns (about a 50x efficiency difference there).