创建zfs 池 zfs_raid0为名称池名称，可以修改

zpool create zfs_raid0 /dev/sda /dev/sdb

查看zfs池是不是生效

zpool status zfs_raid0 

创建挂载点

mkdir -p /mnt/zfs_raid0

挂载zfs池到挂载点

zfs set mountpoint=/mnt/zfs_raid0 zfs_raid0

PVE中挂载
选择此项

填入

添加成功

查看是否挂载成功

 df -Th

Filesystem Type Size Used Avail Use% Mounted on
zfs_raid0 zfs 462G 128K 462G 1% /mnt/zfs_raid0

修改ARC ram 缓存大小 nano /etc/modprobe.d/zfs.conf

添加以下

options zfs zfs_arc_max=1073741824  #单位为bytes，此设置为最大1G
options zfs zfs_arc_min=536,870,912  #最小500MB

更新系统

update-initramfs -u -k all

重启

reboot

重启后检查

cat /sys/module/zfs/parameters/zfs_arc_min
cat /sys/module/zfs/parameters/zfs_arc_max
arcstat
arc_summary | more
arc_summary -d | more

测速
写1G

dd if=/dev/zero of=/mnt/zfs_raid0/testfile bs=1G count=1 oflag=direct 

读1G

dd if=/mnt/zfs_raid0/testfile of=/dev/null bs=1G count=1 iflag=direct

zfs格式化的硬盘重装系统需要先把zfs标签删除才可以重装系统

zpool labelclear -f /dev/sdX 

zpool destroy poolname

wipefs -a /dev/sdX

非zfs可以一次删除多个盘属性，如ext4, btrfs等，但是zfs需要一个个删

wipefs -a /dev/sda /dev/sdb /dev/sdc

一键自动删除zfs脚本

for d in $(zpool status -P poolname | awk '/\/dev\//{print $1}' | sort -u); do zpool labelclear -f "$d"; done

如果知道盘符路径脚本

for d in /dev/sdX /dev/sdY /dev/sdZ; do zpool labelclear -f "$d"; done

优化

nano /etc/modprobe.d/zfs.conf

options zfs zfs_arc_max=2147483648
options zfs zfs_dirty_data_max_max=4294967296
options zfs zfs_vdev_async_write_max_active=32
options zfs zfs_vdev_sync_write_max_active=16
options zfs zfs_dmu_offset_next_sync=0

应用并重启

update-initramfs -u && reboot

zpool set autotrim=on rpool
zfs set compression=lz4 rpool
zfs set atime=off rpool
zfs set recordsize=16K rpool
zfs set sync=standard rpool

rpool为zfs默认安装名称，需要修改为实际阵列名称如zfs_raid0

优化说明

1. Kernel Module Parameters in /etc/modprobe.d/zfs.conf

These are loaded before ZFS starts.

Parameter	Purpose	Effect	My Comment
zfs_arc_max=2147483648	Sets max ARC (RAM cache) size to 2 GB.	Limits RAM usage for ZFS cache.	Be careful — in PVE, ARC is crucial for performance, especially read-heavy loads. If you have >8 GB RAM, 2 GB may be too small and will reduce performance. I’d only set this low if you are short on RAM or running memory-heavy VMs.
zfs_dirty_data_max_max=4294967296	Max dirty data (unwritten data in memory).	Bigger = more write buffering.	Can speed up sequential writes, but increases data loss risk if power fails (unless you have UPS). 4 GB is reasonable for SSDs.
zfs_vdev_async_write_max_active=32	Max active async write I/O per vdev.	Higher can improve SSD/HDD write throughput.	On SSDs, 32 is fine; for HDD pools, too high can increase seek latency.
zfs_vdev_sync_write_max_active=16	Max active sync write I/O per vdev.	Affects sync=always writes.	Could help in NFS/iSCSI workloads; otherwise won’t change much.
zfs_dmu_offset_next_sync=0	Changes how ZFS calculates next sync offset.	Rare tweak, usually left default.	This is an old tuning — in modern OpenZFS, gains are minimal. I’d skip it unless you know why it’s needed.

2. ZFS Pool/Dataset Settings

These are applied after pool creation.

Command	Purpose	Comment
zpool set autotrim=on rpool	Enables TRIM for SSDs.	Good for SSD pools. For HDD-only pools, this does nothing.
zfs set compression=lz4 rpool	Enables lightweight compression.	Recommended — often improves speed by reducing I/O size.
zfs set atime=off rpool	Disables access time updates.	Good for performance unless you need last-access times for files.
zfs set recordsize=16K rpool	Changes block size for files.	Caution — ideal recordsize depends on workload. 16K is good for databases/VM disks, but bad for large media files (hurts sequential throughput). Default 128K is better for big files.
zfs set sync=standard rpool	Sync write behavior.	Standard = safe. Setting sync=disabled boosts speed but risks data loss on crash — fine for scratch data, not for critical VMs.

pool-level, dataset/VM-level, and system-level optimizations:

1. Pool-Level Optimizations

These affect the entire pool.

Action	Why	Command
Enable ashift=12 or higher when creating pool	Ensures writes align with SSD/NVMe sector size (4K). Avoids read-modify-write penalties.	Only set at pool creation: zpool create -o ashift=12
Enable autotrim=on	Keeps SSD/NVMe performance stable over time.	zpool set autotrim=on rpool
Compression=lz4	Often improves speed on fast CPUs because less data is written.	zfs set compression=lz4 rpool
Disable unused features	Example: dedup — uses huge RAM, usually slows down.	Make sure dedup=off

2. Dataset/VM-Specific Tweaks

Instead of changing defaults for everything, tune based on usage.

Workload	Setting	Command
VM disk images	Small blocks reduce wasted I/O	zfs set recordsize=16K rpool/vm-disks
Large media/archive files	Large blocks = fewer IOPS needed	zfs set recordsize=128K rpool/media
Databases / sync-heavy writes	Consider logbias=throughput	zfs set logbias=throughput rpool/db
Non-critical scratch data	Can disable sync for huge speed boost	zfs set sync=disabled rpool/temp

3. System-Level / ZFS Module Tunings

These are safe for NVMe + 16 GB RAM.

/etc/modprobe.d/zfs.conf

options zfs zfs_arc_max=8589934592   # 8 GB

• Use half your RAM for ARC

options zfs zfs_dirty_data_max_max=4294967296  # 4 GB

• Allow larger write buffer

options zfs zfs_vdev_async_write_max_active=32
options zfs zfs_vdev_sync_write_max_active=16

• Boost I/O parallelism for NVMe

Then:

update-initramfs -u && reboot