Property Name | Value | Description |
1. yarn.nodemanager.resource.memory-mb |
55296(54g) |
Total: 62.8g 54g for nm 8.8g forsystem |
2. yarn.scheduler.minimum-allocation-mb |
2304(2.25g) | 1# / 3#: 54g/24core |
3. yarn.nodemanager.resource.cpu-vcores | 24 | 1.5*16cpus |
4. mapreduce.map.memory.mb |
6144(6g) | |
5. mapreduce.reduce.memory.mb | 8192(8g) | |
6. mapreduce.map.java.opts |
-Xmx5120M | < #4 |
7. mapreduce.reduce.java.opts |
-Xmx6144M | < #5 |
The memory of containers, nodemanager, map task, reduce task should based on system memory, cpu numbers.
There are 62.8g memory, 16cpus in nodemanager nodes. The memory could be like following.
Notice: Mostly, 4# = 2*2#, 5 = 2*4#. The following configurations are based on our mapreduce jobs.