Qlik Community

Qlik Scalability

Announcements
Members are not receiving notifications from the community. A bug has been identified and a fix is coming soon.
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
Contributor III
Contributor III

I would like to know everything about RAM recommendations

Hi.

I've searched and found useful documents regarding virtualization, but I can't tell if they are not outdated and still I'm left with some questions.

(This post and this post, have relevant info, and I attach the documents from those threads.)

A) Say you have a physicall server with 32GB of RAM (16GB x 2), if you add one more card of 2GB, would it be better? (2GB more but it isn't balanced).

B) What about doing that on a virtualized server. You assign 32Gb and then jump it up to 34GB or 35GB... would that help?

C) Is there a documment about performance tests with different RAM amount? (or a document without test but explanation to help decide)


D) Do all hardware recommendations apply equally to QlikView 11, QlikView 12 and Qlik Sense?


Regadrs.

1 Solution

Accepted Solutions
Highlighted

Hi Roberto,

  1. To optimize RAM on a server, you should follow the guidelines from the manufacturer since it depends on the architecture used. Generally, maximum memory performance is achieved by EVENLY distributing total desired memory capacity across all operational channels. So adding one small DIMM, will normally break the performance.
  2. A virtualized server will utilize the HW underneath, so it can maximum perform as good as the HW is configured. 
  3. Adding RAM to a server will only help your performance if this is the bottleneck for your performance. If the used RAM reaches the working set (default 70%), the engine will start purging the cached results. This will take CPU time and will impact performance if the CPU is doing calculations at that moment. Also the cached results will be thrown away and thus if the same selections are made, these calculations need to be done again. So you need to take a look at your own case to see if having more RAM can improve your performance.
  4. All HW recommendations are equal for QlikView 11, 12 and QlikSense.


Regards,

Frederic

View solution in original post

5 Replies
Highlighted
Contributor III
Contributor III

No one? Nothing? A link or something?

Highlighted

Hi Roberto,

  1. To optimize RAM on a server, you should follow the guidelines from the manufacturer since it depends on the architecture used. Generally, maximum memory performance is achieved by EVENLY distributing total desired memory capacity across all operational channels. So adding one small DIMM, will normally break the performance.
  2. A virtualized server will utilize the HW underneath, so it can maximum perform as good as the HW is configured. 
  3. Adding RAM to a server will only help your performance if this is the bottleneck for your performance. If the used RAM reaches the working set (default 70%), the engine will start purging the cached results. This will take CPU time and will impact performance if the CPU is doing calculations at that moment. Also the cached results will be thrown away and thus if the same selections are made, these calculations need to be done again. So you need to take a look at your own case to see if having more RAM can improve your performance.
  4. All HW recommendations are equal for QlikView 11, 12 and QlikSense.


Regards,

Frederic

View solution in original post

Highlighted
Contributor III
Contributor III

Thanks!

Just to close B (and all of them with it)... So if the HW underneath performs great (with evenly distributed RAM, say 8 cards of 16GB) one can virtualize a server with 33GB of RAM and expect a better performance than having it set up at 32GB?

Regards.

Highlighted

Yes, in theory the VM with 33GB will perform 'better' than the one with 32GB. But there are many configuration parameters with different VM's so there might be difficulties because of that. Also the performance gain of adding 1GB will be very very small: it's only adding 3% more memory. Eg. if your RAM fills up after 1 hour of extensive use with 32GB, it will probably fill up in 1hour and 2 minutes with the 33GB, so there is not much of a gain.

Best regards,

Frederic

Contributor III
Contributor III

Yes I know 1GB increment is almost nothing, which is why I chose to use it in the question, if it's true in the most extreme case, then it's true in most obvious ones (incrementing 16GB/50%).

Thank you very much, you've given me valuable information.