Anita,
As the previous comment states it's difficult to provide specifics due to the myriad of different scenarios that large datasets can throw up, therefore I'll recommend a few generalisms that are more 'best practices' for most QV applications - it's just you don't NEED them normally whereas with larger models they're crucial.
Firstly attached are several documents that cover the subject of optomization and scalability which should give you the general pricipals need to tune any app you create.
From my own experience I'd point out the following:
- A large amount of data is going to be...large; there's no getting away from that we can only optomise and reduce so far so any hardware will have to be specced accordingly.
- Use autonumber() round your key fields as these take up alot of memory by their unique nature.
- Bear in mind Qlikviews compression method: 'Column Compression': 1million rows of 'Salutation' (Mr, Mrs etc) will take up far less memory than say 1milion rown of telephone data as Qlikview replaces each unique value it finds with a token and the fewer tokens the less memory is used. Therefore look to split things like phone numbers into 2 columns that way you get good compression on the Area code at least.
- Ensure the document is set to be 'Preloaded' into the servers memory after each reload.
- Don't be affraid to limit access to chart objects until a certain minimum level has been reached as Objects only take up memory when 'drawn' - ie in the objects 'Calculate Condition' box use something like 'getpossiblecount(My_Field)<100000'.
- You should use some form of multi-tiered architecture whereby a QVW extracts the data and generates QVD's for subsequent dashboards to quickly and easily load from; this stops say 3 similar dashboards that may take 3hrs each to load hitting your source DB in you reload window of 7hrs - have the 'mule' do it once in 3hrs then have the dashboards load from the QVDs in 3x 10mins.
- Once you get really large you may need to look at clustering.
Hope all that helps - let me know how you get on. - good luck,
Matt - Visual Analytics Ltd
.png "Luminary Alumni"){kind=icon}
