7 Replies Latest reply on Jan 30, 2015 7:17 PM by dng

    AF Rollup and DataLink

    alexandre.bouffard

      Hi,

       

       

       

      I need to rollup attribute values dynamically (on demand - without using a PI tag) and I understand that Asset Based Analytics does not have this feature.

       

      I could use the Rollup custom reference but RDA (Rich Data Access) is not implemented so it does not work with DataLink for instance.

       

      So I thought I could implement RDA by adding the SupportedDataMethods implementation provided in the white paper.

       

       

       

      I just wanted to make sure that I am on the right track and I didn't miss anything (eg. Rollup custom reference with RDA implementation already available somewhere).

       

       

       

      Thanks,

       

      Alexandre

        • Re: AF Rollup and DataLink

          Hello Alexandre,

           

          Alexandre Bouffard

          without using a PI tag

           

          What's the background of this requirement?

           

          Please note that we like to support you using PI and not with preventing you from using PI.

          • Re: AF Rollup and DataLink
            dadijke

            You can build your own AF Data Reference to make use of Rollup. A white paper + example is available in the Library (White Papers > PI AF > Implementing AF 2.x Data References). You can make your own modifications to this Data Reference rollup to fit your needs.

              • Re: AF Rollup and DataLink
                alexandre.bouffard

                @Gregor

                 

                I work mostly with backfilled data and occasionally time series (and associated AF Elements) are added or modified.

                 

                Also the dataset is quite large and rollups are not systematically used afterwards therefore for some use cases such as this one it is quite acceptable to deal with a slightly slower access time in order to access up to date aggregations for past and present time.

                 

                @Danny

                 

                Yes this is the white paper I was referring to, thanks.

                  • Re: AF Rollup and DataLink
                    skwan

                    Alexandre:

                     

                    In your situation, where you're rolling up a large number of values from a large dataset, please consider the effects that you may be causing on your clients and servers if you choose to implement a client side rollup application.  As you are probably aware, the rollup data reference example we provided on vCampus and also the white paper perform the actual rollup calculations on the client side.  Therefore, every time you perform one of these operations, the data reference must first figure out which data it needs, perhaps needing to traverse the hierarchy to find all the attributes that are involved in the rollup, go get the data from the sources, perform the calculations, then presenting the result.  On such a client side operation, you must take great care in optimizing the operations, for example, building a list, perform bulk calls, perhaps asynchronously, etc.  Otherwise you will find that the performance may not be satisfactory to you.  Just take a very simple example, you want to rollup 10,000 attributes.  If you try to retrieve the values in a loop, you're going to have network latency * 20,000 overhead.

                     

                    One of the main objective of AF 2014 with Asset Based Analytics is to help minimize the scenario above by "pre-calculating" values so that if you had to do a client side calculation, you can do it with the pre-calculated results rather than having to retrieve every single value every time.  As an example, let's say you configure some analyses with AF 2014 that runs on a periodic schedule of 5 seconds and you store the results in PI tags.  Now if you wish to do a client side rollup, you can rollup these (pre-calculated) results rather than the raw values.  Since you have fewer values to retrieve, your performance will be greatly improved.

                     

                    Please also view the vCampus Live! 2012 presentation by Ray Hall and vCampus Live! 2013 presentation by Chris Manhard on ways to optimize AFSDK applications.  These are available for download from the download center.

                      • Re: AF Rollup and DataLink
                        alexandre.bouffard

                        Thanks Steve for the insights.

                         

                        Actually what I meant by a large dataset on which rollups are not systematically used is that for let's say 100,000 attributes, I could have 20,000 rollups of 5 attributes each and only a tiny fraction of those 20,000 rollups would be read.

                         

                        In other words they are just lots of simple formulas with a small number of parameters each but those parameters cannot be defined explicitly since they are based on element hierarchy.

                         

                        I have been using the rollup custom data reference for a while in System Explorer and Coresight and the performance is adequate. My question was coming from the fact that the white paper implementation cannot be used as is in DataLink since it does not implement RDA.

                    • Re: AF Rollup and DataLink
                      bjsower

                      Where can I find the source for the examples that go with this white paper?  Having difficulting finding it with the transition from vCampus to PI Square Community.