Modifications to make on large node? (1TB RAM)

Mirkat

New Member
Jun 17, 2021
1
0
1
31
I am wondering what is some of the changes I should make to KSM, swap and anything else you guys suggest for very large RAM nodes?

Currently I have read I should make the following changes to KSM

Code:
KSM_THRES_COEF=10
KSM_SLEEP_MSEC=10

The only problem I have found with this so far is the total "KSM Sharing" stays around 100GB but with the default value it goes as high as 300GB should I switch that back?

The server also has ZFS set to 20GB min 40GB max

Also I have changed swap to enable at <10% RAM

Code:
vm.swappiness=10

I have been running tests with swap. I eat 200GB of the nodes RAM during a test and found that

Code:
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root     65815 21.5  5.6 69967176 59383812 ?   Sl   Jun10 2164:55 /usr/bin/kvm -id 16

This virtual machines RSS memory was lowered which is great as it's reserving 60GB RAM but was only currently using about 3GB on the graphs.

Code:
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root     65815 21.4  4.5 69967176 47985040 ?   Sl   Jun10 2166:10 /usr/bin/kvm -id 16

What happens to this 199GB swap now it's been hovering around that now for hours

Few other changes added are

Code:
vm.min_free_kbytes=524288
vm.nr_hugepages=72
vm.max_map_count=262144
vm.overcommit_memory = 1

2955c912108ddffbdb2d91bf05165121.png

Before/After eating RAM



a5eb8a51c958e75819f5ee1524880478.png
 
Last edited: