You were correct to ask about lz4 in the context of GELI.
lz4 is not usually a problem with spinning disks because the disks can't write fast enough for the lz4 to slow them down.
However, the very latest NVMe SSDs on the very latest PCIe bus variants can, almost certainly, write a lot (ballpark 10 times) faster than "ordinary" CPUs can compress data for writing with lz4.
Note that GELI
will slow down your disk access, potentially quite a lot, which is why it's essential to thoroughly investigate the
aesni(4) module.
This post is 10 years old, but might be a good place to start.
Once you know how much throughput the GELI provider is costing, you can consider the cost of lz4 in the ZFS layer (because the lz4 compression will only have to keep up with the GELI provider)..
So it's a complex ecosystem, impacted by CPU, AES-NI acceleration, PCIe bus, M.2, and SSD hardware; plus the GELI and lz4 software layers.
There is no simple answer if you have very fast SSDs.
Bottom line: if you have installed and configured GELI, and are not sure about lz4, it's not hard to enable lz4 selectively and run a disk I/O benchmark.
Do remember the old axiom that "90% of applications spend 90% of their time reading (or waiting to read)", and lz4 is
really fast at decompressing. So, although writing will be slower, it's rarely the bottleneck.