Performance Issues With 8 Socket Hardware

For larger QlikView applications the required data for aggregations is held in memory through the CPU cores to perform calculations.

The number of CPU core and the speed of the CPU cores can lead to extensive demand on the memory bandwidth and contention for resources can occur. This can significantly slow down the processing of the results.

Qlik Scalability Center's research shows that current 8 socket CPU's architecture cannot connect all CPU sockets to each other, they have to go through the memory and maybe even a buffer. This means that additional hops are required for a CPU to reach the memory. The below link provides more details the 8 socket architecture and it natural impact on calculation performance for software with high CPU utilization like QlikView and Qlik Sense. 

http://www.qdpma.com/systemarchitecture/systemarchitecture_qpi.html

Performance degrade due to 8 socket hardware has been seen for:

• HP-DL980  
• IBM X3850 X5 

A general guideline on performance is that, larger QlikView applications perform better on a smaller number of faster CPU cores with a faster memory bus can perform better than a larger number of slower CPU cores. Conversely, where smaller QlikView applications are used, a larger server with more CPUs and slower CPU cores can support more users than a smaller server with faster CPUs.

The recommendation is to use white listed hardware for QlikView or Qlik Sense deployments. These recommendations are based on extensive test and evaluation by Qlik Scalability Center. 

Server Hardware Whitelist From Qlik Scalability Center

For accurate deployment advise, please contact your local Qlik sales team for scaling and deployment recommendations.