Random performance

[ramonarcusa36]

Hi everyone.

I'm one of those guys who's having performance issues and came to this community searching for ideas to improve my application. I've read a lot around here, specially this thread, which is great. I got some ideas, I simplified some expressions, and my application's performance got slightly better.

The green progress bar that frequently stayed for a while on the bottom right corner of my charts, now it's hardly shown. But it keeps showing the hourglass (well, you know, hourglass looks like a blue spinning circle now ) for some seconds after I make any selection and before calculating my chart.

And there's still something weirder (the main reason of this new discussion I'm opening): the performance problems are quite random. I mean, why does it take to QlikView just one second to perform a calculation, and eight seconds to make the next one? Did it go to sleep and has to wake up or something? Bear in mind that both operations are pretty similar, I'm just making two selections on two "sibling" fields, both coming from the same table, with the same (approx.) number of different values.

I know this is a quite vague question but, could you guys point out some reasons that could be responsible of such randomness? This is just driving me crazy, because I don't know how I can optimize something that takes 1 second to calculate now, 6 seconds later, 1 second afterwards....

Just to clarify: I always test by making consecutive selections in two or more different fields, because if I take just one field for my tests I know it would be quite faster after the first selection, because of caching operations and all those things.

Thanks in advance!

[jerrysvensson]

Are you running Publisher and QlikVIew server on the same machine?

[ramonarcusa36]

No, we're not running Publisher in the server or in any other machine... we don't use Publisher.

Thanks for your answer.

[jerrysvensson]

You should buy a Publisher.

Do you have any virusscanning activated on the machine? That will have an effect on performance when that runs.

[ramonarcusa36]

Maybe we should buy it, I agree with you... but I think that decision isn't up to me, you know...

I don't know if there is a virus scan running, I will check it with the guys who manage that...

For testing, I'm rebuilding the official document from scratch, copying everything step by step to a new document, trying to see which new feature makes it slower. So far, I've created a new document, with the same load script, one simple table with all expressions copied from the original document, some multibox selectors... No matter how many things I add to the new document, it always shows a very better performance than the first one. The official document was built with v10, and the reconstruction is being done with v11, which is our current version (for everything: client and IE Plugin). Do you guys think that having built our official document with v10 could be one of the reasons of its worse performance?

I know I'm not talking about random performance in my previous paragraph, but as I said before, our performance went slightly better, but we're still not happy with it.

Thank you all.

[ergustafsson]

Hi Ramon,

Sounds like one calculation is cached, while the other one is not. Have a look at http://community.qlik.com/blogs/qlikviewdesignblog/2014/04/14/the-qlikview-cache for more information.

Let me know if this helps.

Regards,

Erik