topic Re: Windows 10 setup for better performance in QlikView

Windows 10 setup for better performance

pedromsouza — Fri, 05 Feb 2016 13:53:27 GMT

Hi,

I'd like to know how to setup Windows 10 for better Qlikview performance (virtual memory and other tweaks). I searched about it and find information about Qlikview Server, but I want info about tweaks for Qlikview desktop (for development).

Recently I edited a QVW that would freeze Windows while reloading. It happens when it hits 100% ram usage.

I have a laptop with i7, 8gb of ram (but video card takes 0,5).

Regards,

Pedro.

Re: Windows 10 setup for better performance

Colin-Albert — Fri, 05 Feb 2016 17:57:04 GMT

Try debug load and load fewer rows say just 100 using the Limited Load option, and check for synthetic keys and tables that are not associated.

If the data model is OK, then install more RAM! QlikView is memory hungry!

Re: Windows 10 setup for better performance

HirisH_V7 — Fri, 05 Feb 2016 18:07:13 GMT

Hi,

You can do one thing when it freezes, if you hit esc (closing script editor), does QlikView come back alive!

Increasing Ram Helps.

-Hirish

Re: Windows 10 setup for better performance

pedromsouza — Sat, 06 Feb 2016 19:12:46 GMT

I think the problem is related to 3 master calendars with auto increment functions to fill the days without "value".

Re: Windows 10 setup for better performance

pedromsouza — Sat, 06 Feb 2016 19:13:23 GMT

Thanks! Will try that... I forced reboot my laptop last time.

Re: Windows 10 setup for better performance

pedromsouza — Sat, 06 Feb 2016 19:14:07 GMT

You think increasing virtual memory won't help, then? Thanks.

Re: Windows 10 setup for better performance

oknotsen — Sat, 06 Feb 2016 21:46:43 GMT

The moment Windows need to start swapping memory (used by QlikView) to disk (in other words "use virtual memory"), you can basically throw it out of the (real) window. Speed drops to next to nothing.

100% CPU usage is not weird.

Re: Windows 10 setup for better performance

oknotsen — Sat, 06 Feb 2016 21:48:42 GMT

Multiple master calendar that results in a bunch of synthetic keys, that will eat up memory.

The wildest single master calendar will just be a few thousand rows and as a result will not eat up loads of memory.

Re: Windows 10 setup for better performance

petter — Sun, 07 Feb 2016 09:20:59 GMT

If you run out of RAM during or at the end of a reload you can in order of priority:

1) Reduce the data you loading

2) Make sure that the footprint of the data you are loading becomes less in QlikView

3) Buy more RAM

For 1 and 2 you should first of all find the real culprit of the problem. Is it a single table? If so analyze this table in the context of the points in 1 and 2.

1) Reduce the data or the footprint of the data you are loading:

a) Do you need all the rows you are loading in every table? Could you analysis application really be split into

several apps with different year-spans or quarters or something similar.

b) Do you need the full granularity of the rows? Could you summarize from minutes into hours or days instead?

c) Do you need all the columns of every table?

2) Make sure that the footprint of the data you are loading becomes less in QlikView

a) For every column that will be turned into fields could you decrease the number of distinct values. By

for instance rounding of a sales transaction date from full resolution into a day-resolution? It is a

similar operation as b in point 1 but not the same. Here you would reduce the cardinality of the field

itself whereas in b you would reduce the number of rows for a table. This could have a tremenduous

impact on reducing the footprint of this field alone. It is due to the fact that you would allow QlikView

to use it's behind the scenes symbol tables to store way less data - a form of compression that often

saves a lot of RAM and consequently also speeds up any calculations.

b) Make sure that you don't allow for synthetic keys that are not necessary. In most situations synthetic

keys are unnecessary - so try to avoid them at all costs. They can consume a lot of RAM and be very

costly when it comes to performance.

c) Analyze the number of tables and how they relate to each other. Is it a simpler way of getting them to

relate? Will it have an impact on the footprint? Do you really need all of the tables? Could some of them

be merged or joined together.

3) Buy more RAM:

This might not even solve your problem. So at least before doing such a thing make sure that you load half the data and see how much RAM you have left after that. If you don't get to load even the half it seems to suggest that even doubling your RAM will still leave you in trouble....

Mind you this is not an exhaustive list of what you can do when you run into these kind of issues. I wrote this out of the top of my head...

Re: Windows 10 setup for better performance

Colin-Albert — Sun, 07 Feb 2016 17:45:53 GMT

Does the data load OK if you just load one master calendar?

Re: Windows 10 setup for better performance

pedromsouza — Thu, 11 Feb 2016 21:55:17 GMT

No, it freezes.

Re: Windows 10 setup for better performance

pedromsouza — Thu, 11 Feb 2016 21:57:23 GMT

Thanks. Very helpful answer. I removed some unnecessary columns, but it still uses all 8 GB of RAM. I think it's related to the master calendars (3).

Re: Windows 10 setup for better performance

pedromsouza — Thu, 11 Feb 2016 21:58:59 GMT

I changed all calendar columns names to avoid synthetic keys, but it still uses 100% ram and freezes. But I'm almost certain the issue is related to the calendars.

Re: Windows 10 setup for better performance

petter — Fri, 12 Feb 2016 06:01:11 GMT

Can you descibe your master calendars? The structure, how many columns in each and which columns? How rows do they have each? If you share a screenshot of the Table Viewer we could get a better understanding of your data model and possibly give you some advice on that.

Re: Windows 10 setup for better performance

Colin-Albert — Fri, 12 Feb 2016 09:30:02 GMT

Can you enable debug mode and just load 10 rows into each table using Limited Load. This will allow you to see how your data model looks and if you could post a screen-shot of the model this may help identify the problem.