topic Re: Do we have linear relationship between RAM and performance in customer facing nodes? in Management & Governance

Do we have linear relationship between RAM and performance in customer facing nodes?

Huiying — Mon, 25 May 2020 08:46:21 GMT

I understand Qlik is in-memory technology. Does it mean that adding more RAM can always help performance?
Now we reached bottleneck with 2 customer-facing node (350G RAM each). Cost wise, adding more RAM is better than adding more nodes. Is there anything else to be considered before deciding more RAM or more nodes?

Re: Do we have linear relationship between RAM and performance in customer facing nodes?

Anders_Eriksson — Mon, 25 May 2020 09:03:13 GMT

You should do a simple investigation what causes the bottleneck.
Is it RAM or CPU or some other cause like network or disk access?
More RAM is good but won't help if the bottleneck is caused by something else.

Re: Do we have linear relationship between RAM and performance in customer facing nodes?

Huiying — Mon, 25 May 2020 09:55:07 GMT

Thank you for replying!
At this moment, the issue seems mainly in RAM, as servers are constantly over low working set.
I guess there is no upper limit how much RAM we can add, right? And probably there is no negative impact if we add RAM to 700+G for each server?

Re: Do we have linear relationship between RAM and performance in customer facing nodes?

Anders_Eriksson — Mon, 25 May 2020 10:26:45 GMT

Servers will not go under lower limit unless they have to.
The server being over the lower limit is not a problem but per design.
It is the upper limit that is critical for RAM.
Still advice looking into and making sure it is memory that is the problem and not something else.
Don't know of any limitation in memory size in Sense as long as the Windows server can handle it.

Re: Do we have linear relationship between RAM and performance in customer facing nodes?

Huiying — Tue, 26 May 2020 07:18:26 GMT

Thank you! I'll investigate this week.