ARM64 Linux Performance Tuning

This is personal notebook for ARM64 performance tuning.

NVME 4K Write Performance

Situation:
Run fio 4K write on 24 NVME drivers, but some drivers might not get higher IOPS performance

Solution:
It should be caused by enable SMMU, so, if doesn’t use VM, just disable iommu to get better performance
Add iommu.passthrough=1

General IO performance Tuning

For high bandwidth card, most time it needs to disable irq_balancing
$ systemctl stop irqbalance.service

Some Performance Tuning Parameters

ulimit -s unlimited 
ulimit -l 2097152 
echo 8 | sudo tee /proc/sys/vm/dirty_ratio 
echo 1 | sudo tee /proc/sys/vm/swappiness 
echo 1 | sudo tee /proc/sys/vm/zone_reclaim_mode 
echo 3 | sudo tee /proc/sys/vm/drop_caches 
echo 0 | sudo tee /proc/sys/kernel/randomize_va_space 
echo 0 | sudo tee /proc/sys/kernel/numa_balancing 
echo never > /sys/kernel/mm/transparent_hugepage/enabled 
cpupower frequency-set -g performance 
export MALLOC_CONF=thp:always