Hi Dess,

I am a bit confused. Lets me recap our conversion:

1. Me: The Paste-method of the RadGridView is slow when pasting half a million records. This is due to the StringTokenizer.
2. You: RadGridView is not really able to handle half a million records, try the RadVirtualGrid.
   
   (Me: Taking days to replace the RadGridView by the RadVirtualGrid. After implementing, pasting half a million records. Still slow.)
   
   
3. Me: RadVirtualGrid is also slow, because it also uses StringTokenizer.
4. You: Of course it is slow, you are pasting the same amount of records.
5. Me: Eh... then why suggesting to use the RadVirtualGrid while pasting half a million records was the problem in the first place?

Please see these figures:

What I do not understand it this:

1. As per your own documentation RadGridView should support a million records. It doesn't.
2. Than you give me the suggestion to transfer to RadVirtualGrid, because that would support a million records. It doesn't.
3. You say "The StringTokenizer is globally used in the Telerik Presentation Framework and the TextPrimitive". So it is a core component, you how great would it be if that component is really fast?
4. I gave you a working patch (see the numbers). You answer with "the RadVirtualGrid case will be considered for improvement when addressing this request",but... the problem is not with the RadVirtualGrid or the RadGridView, the problem is with the StringTokenizer, one of your core components.
5. In short: I do not really understand the hesitation...

Hi Dess,

Maybe in the case of RadVirtualGrid you might consider this a bug. In case of RadGridView it was NOT a bug, since RadGridView was not created for millions of records. RadGridView is and pasting a million records should not take minutes... right? :)

Regards,

Martin

Hi Dess,

You might have noticed I am switching to RadVirtualGrid. I discovered that RadVirtualGrid is also using StringTokenizer. This means that the same performance problem will remain.

Regards,

Martin

About performance in general: What may we (clients) expect? This article states "You want to bind RadGridView to over a million items. In such cases it is recommended to use Virtual Mode". But... we are not even half of that and already are some operations far too slow.

So can you express some minimum expectations? Something like: When you load a RadGridView with 1M records, all bulk operations (like loading, select-all, delete-all selected, copy, paste, etc.) should take a maximum of 10 seconds. (Maybe even  based on a certain processor type).

In that case we can report bugs instead of requests. In that case we also know when to switch to Virtual Mode or RadVirtualGrid. In that case I also expect Progress to test the performance before a new release. 

This feature requests is just one example of many about performance issues of RadGridView. Yesterday I also patched a piece of code to increase performance for a bulk append operation. Right now, for me, it is frustrating to do so, because I have the feeling I am fixing the problems of Progress. But if I am expecting more of the grid than what is defined, then it is my choice and my struggle. 

When I create a new request/bug, then I see the bar as you have shown. 

When I reply (or add a new comment), I see this bar:

Hi Dess,

I'll do better than that and focus on the component in question: The StringTokenizer. See the code below (off topic: When create a new request or bug, I can insert code. When I am answering on an existing request/bug, that button is gone... why?)

In this example I create "clipboarddata" with 50,000 records and 4 columns. The total length of the string is 1.2MB. Constructing this tokenizer takes 38 seconds (on my machine). This delay grows exponentially to the size of the data! Using my patch it takes only 0.03 seconds and grows linear. You can imagine how slow it performs having clipboard data of 10MB.

This tokenizer is used during pasting plain text and pasting HTML (and maybe in other places). 

==================================================================

static void MeasurePerformance()
{
var sb = new StringBuilder();
for (var i = 0; i < 50000; i++)
sb.Append($"{i}\t{i + 1}\t{i + 2}\t{i + 3}\r\n");

var clipboardData = sb.ToString();

var stopwatch = new Stopwatch();
stopwatch.Start();
var tokenizer = new StringTokenizer(clipboardData, "\n");
stopwatch.Stop();
Console.WriteLine(stopwatch.Elapsed.TotalSeconds);
}

Addition:

As suggested in my first post, StringTokenizer can make use of String.Split if the delimiter is only one character. But... In the methods MasterGridViewTemplate.GetHtmlData and MasterGridViewTemplate.GetTextData why not use String.Split and remove the use of the StringTokenizer?