General Considerations When Selecting Server Hardware
The purpose of this technical note is to provide starting guidance on what to consider when selecting hardware for the Qlik Associative Engine in Qlik Sense® and QlikView® deployments.
Server selection for any given customer should be based on recommendations and tests provided by trained Qlik professionals. For information on how to contact Qlik's Professional Services, see How and when to contact Qlik's Professional Services and Consulting Teams.
Processor
In current generations (2023) of processors, there is not much difference between AMD and Intel.
Memory
More RAM allows more result sets to be cached, which typically results in better performance.
The Qlik Associative Engine allocates memory uniformly over all available RAM, which means that it is important to always have the amount of memory per CPU socket evenly distributed.
For information on the memory configurations for best performance (including which RAM configurations can be deployed whilst keeping the bus speed as high as possible), contact the server manufacturer.
Architecture
To compare the performance of different servers, you can use the theoretical calculation capacity:
Theoretical Calculation Capacity=#sockets ×#CPU cores×base clock speed
This gives the theoretical number of calculations that a server can do every second. Servers with higher theoretical calculation capacity generally perform better than servers with lower theoretical calculation capacity.
New architectures usually improve on the number of instructions per clock cycle and thus provide improved performance in comparison to older architectures.
Sockets
Since the Qlik Associative Engine is good at using memory as well as the processing capacity of all CPUs, the chipset architecture must provide fast communication between the CPU sockets to keep the latency towards the memory as low as possible. Chipset architectures that allow for direct connections between all available CPU sockets via high bandwidth links are therefore preferred.
- Single-socket configurations: Single-socket configurations perform very well as all memory is directly connected to the CPU.
- Two-socket configurations: These configurations perform very good providing there is enough bandwidth between the CPUs and towards the memory. For example, Intel CPUs with two or three processor interconnects perform better than those with just one. Using a single Intel CPU in a two-socket configuration also provides good performance.
- Four or Eight-socket configurations: In most situations, it is not advised to use these types of configurations with Qlik Sense or QlikView as they more frequently introduce latencies to remote RAM.
Core Count and Clock Speed
It is important to find a good balance between the core count and the clock speed. Even though the Qlik Associative Engine is very good at multi-threading and uses all cores available to it, some procedures are still single-threaded and run faster with higher clock speed.
BIOS and Operating System Settings
Server BIOS and operating system settings can have a significant impact on the performance of Qlik Sense and QlikView deployments. For more information on the recommended settings please visit Quick tips #8 - Server settings for best performance.
The attached recommendations are stored for archival purposes. No additional .pdf's will be created. Server selection for any given customer should be based on recommendations and tests provided by Qlik professional services.
