It doesn’t matter if you add columns at run-time or not. The main factor is how many column you are displaying and how many rows will be displayed. And note that is different from the total set of columns that could be displayed and total number of data items:
Example: I have a 10000 person database with 100 different statistics on each person. I set up my DataGrid to have 100 columns representing each statistic, but because it won’t fit on the screen I set horizontalScrollPolicy=”true”. Still, the DG is likely to be full screen so it can show about 20 columns and 30 rows. Note that if do not set horizontalScrollPolicy=”true”, the DG will cram all 100 columns on screen. This makes a significant difference. Instead of 20x30=600 renderers, it is 100x30=3000 renderers. The DG will take more than 5 times longer to set up and render 3000 renderers instead of 600 renderers, memory usage will go way up, and even things like mouse tracking and CPU utilization at idle time will be impacted. The next important thing to remember is that DG has a very generalized update mechanism. If it detects a change to any of the 10000 person records, it will refresh “all” of the renderers on the screen. It does not optimize for whether that renderer is associated with that data item because labelFunctions and custom renderers make it possible for any data item change to affect every renderer. This is very inefficient, but we have to be this conservative otherwise your renderers could miss updates. This is important in how you setup and maintain your dataProvider. A common mistake I see is to take some XML, make a new ArrayCollection, and loop through the XML creating Person records and using push/addItem to add it to the ArrayCollection. Every time you do that, the AC generates a change event, and if the DG is already bound, it is going to try to refresh all of its renderers. So, aggregate changes. If you are converting XML, make an array, convert items and push them on the array and then set the whole array into the ArrayCollection’s source. HTH, On 3/22/10 4:28 PM, "fusionpage" <[email protected]> wrote: Does anyone have any tips or tricks, in general, about optimizing the Flex client for best performance in populating/rendering large datasets in an AdvancedDataGrid? I'm specifically NOT asking about improving the server-side. I'm trying to understand what the factors are on the Flex client-side that impact how fast data loads/renders into an ADG, when the columns use in-line itemrenderers and/or the columns are added at run-time. Are there any significant difference in performance when comparing different design patterns or ways of generating the datagrid? e.g. does it matter if it generated via pure AS3, or if you use MXML? Is performance the same no matter which design pattern one uses? Assume the data is loading via a RemoteObject call. In my case via a ColdFusion CFC. Any tips, tricks, or links so I can learn more? So, far, in my experience, most of the performance issues come from the server-side, but I may be ignorant on the impact of the design on the Flex side. Thanks, Don -- Alex Harui Flex SDK Team Adobe System, Inc. http://blogs.adobe.com/aharui
