What is your proposal:
On the machine with PMEM hardware, we can utilize this hardware capability, migrating BE pods memory to PMEM instead of evicting them directly when the system archives the memory safety threshold.
Why is this needed:
PMEM can be used as the system memory, then we suggest to implement a differentiated memory policy for the machine with PMEM hardware, which can make full use of the hardware capability.
Is there a suggested solution, if so, please add it:
When the memory utilization exceeds the safety threshold, we can choose to migrating BE pods memory to PMEM, BE pods can continue running on PMEM or temporarily freezing the CPU, which can reduce the memory pressure of LSR & LS pods. And when the memory utilization is far below the safety threshold, we can choose to migration BE pods memory back to DDR.
What is your proposal: On the machine with PMEM hardware, we can utilize this hardware capability, migrating BE pods memory to PMEM instead of evicting them directly when the system archives the memory safety threshold.
Why is this needed: PMEM can be used as the system memory, then we suggest to implement a differentiated memory policy for the machine with PMEM hardware, which can make full use of the hardware capability.
Is there a suggested solution, if so, please add it: When the memory utilization exceeds the safety threshold, we can choose to migrating BE pods memory to PMEM, BE pods can continue running on PMEM or temporarily freezing the CPU, which can reduce the memory pressure of LSR & LS pods. And when the memory utilization is far below the safety threshold, we can choose to migration BE pods memory back to DDR.