We are in a very sticky situation with our Qlik setup due to a performance issue.
Based on an organizational decision, it was decided that the enterprise reporting platform move to a new data center which had a different hardware setup to the one that we are using. The Qlik setup was also moved accordingly.
A bit more detail about each environment where each has a pair of Windows 2012 Servers, albeit with different hardware :
Old (2 servers with the same hardware configuration)
2 x E5 2697 v3 (2 sockets, 28 cores, 2.6 GHz)
512 GB RAM
New (2 servers with the same hardware configuration)
4 x E7 8891 v3 (4 sockets, 40 cores, 2.8 GHz)
576 GB RAM
The services are distributed across both the environments in the below order
We are currently running the same set of QlikView dashboards on both the environments pointing to the same database and therefore the data source and data volumes are the same for both environments. Also, the QMC setting on both the environments are as close to being identical as possible. These dashboards are quite complex and hold a lot of data (approximately 50-70 million records). While there is an argument about 'optimizing' them and make them more efficient, the fact is that they are showing extremely disparate results even when the code and the data is the same. The only thing different is the environment and we have tried to go over all the settings/configurations on the new environment and have tried the following fixes:
- Enable Hyperthreading on Server 2 and disable it on Server 1
- Create Antivirus exclusion rules for daily scans for QlikView related files
- Increase the Heap size to run more than 9 engines
The performance problems are both from a dashboard reload perspective as well as from an end user experience.
The Old setup is not only refreshing the dashboards in half the time as the new setup, their experience from a user perspective is also very fast and responsive. The new environments just doesn't seem capable to handle the loads. The extraction from the database is almost the same; it is the transformation steps on the new server (after it has loaded the QVD) which is very slow as compared to the old server, almost 2-3 longer.
Also, when opening the dashboards via the Access Point, the dashboards on the new server seems to take ages to render the dashboards and we can see 100% CPU plateaus making the server unresponsive.
At this point of time, we seem to have run out of ideas and have raised a ticket with Qlik. However, any additional perspective is gladly welcome.