Hardware configuration reference

Configuration description

Each CPU Core can undertake real-time calculation of 20 million rows of data.

If calculated according to 1000 Bytes per piece of data, 20 million pieces of data are almost 20G of original data. Assuming that the compression ratio of 20G original data is 1/3, the data loaded in the memory is 1/3, and the data in the memory is almost 2G. Therefore, when calculating about 20 million pieces of data, the memory consumes about 2G.

The more CPUs configured, the better the system performance.

In the MPP data mart, CPU expansion and performance improvement are basically linearly related. Due to memory-based calculations, memory coordination is required.

When calculating the CPU, you also need to consider the amount of concurrency.

The number of online users and the number of concurrent queries submitted by online users will have different requirements for CPU and memory. We need to ask for the number of queries executed per second in order to estimate the CPU and memory requirements for concurrency. Generally speaking, the demand for concurrent queries will rise slightly. The formula estimated by the industry is generally to add about 5% of memory and CPU to the original basis for each concurrency.

Sufficient memory can reduce data exchange and also help improve performance.

The amount of thermal data for each system is not necessarily, some are 1/5, some are 1/3, and some are 1/10. When the memory is large enough, all the hot data are saved to the memory, and then adding the memory after that will not have much effect on the performance of the system.

In general, the ratio of raw data to the amount of memory can be defined as 10:1, and the ratio of raw data to CPU Core can be defined as 20G:1. In this configuration, the query speed can reach a second-level response.

1T data volume requires 100G memory and 50 CPU Cores. Consider the following configuration: four 2-channel PC Servers, each with 2 CPUs, and each CPU has 6 Core, and each has 24G of memory.

500G data volume requires almost 50G memory and 24 CPU Cores. Consider the following configuration: three 2-channel PC Servers, each with 2 CPUs, and each CPU has 4 Core, each with 16G memory.