0 Replies Latest reply: Jul 12, 2017 5:05 PM by Kyrylo Prykhno RSS

    Performance issues – huge delays on selections

    Kyrylo Prykhno

      Dear All,

       

      I've faced with the following problem during development of middle+ size application:

      When user select the specific value(s) in the list, for example, set of months ( a Qlikviewfield in Master calendar) - there is a delay in 15-20 seconds BEFORE QLIKVIEW STARTS TO RECALCULATE OBJECTS (object recalculation does not consume so much time = 2-3 sec). During this delay, all 16 cores of CPU are loaded up to 100%.

       

      Details: QlikView 12.1 SR4, QVW size – 4Gb, 180M records х 80 fields in fact table, Star Schema (+10 Dic tables)

      Environment (Virtual):  x64 WinServer 2012R2, 128 Gb RAM, 16 CPU (logical) cores AMD Opteron 6174

       

      I’ve done and checked a lot in order to optimize performance:

      1. Star schema
      2. No synthetic keys
      3. Many flags (1/0) are pre-calculated in script and joined to fact table (it is the reason of 4Gb qvw size – there is no lack of RAM, the bottleneck is CPU)
      4. All link fields between tables are numeric (autonumber)
      5. No “if” and “aggr” functions in expressions
      6. No “global” (started with = sign) calculations for variables. Anywhere they are used via dollar-sign expansion
      7. Document Analyzer recommend just remove some unused fields - done

      In addition:

      1. There are many (60) variables with expressions text
      2. Some variables are used in Set Analysis conditions via $()
      3. There are 3 data islands , BUT with Alternative State usage

       

      I realize, the App is not simple and requires much CPU resources for processing, but the most of all I consider of DELAY between SELECTION and the start of object recalculation (EVEN IF THERE IS NOTHING TO RECALCULATE – ONLY LISTS ON SHEET).

       

      Of course I think that hardware tuning could help – physical vs. virtual, Intel vs. AMD, disable NUMA. But not significantly …

       

      Does anybody know what could cause the described delay in the associative model performance?

       

      Thanks in advance!