Comment by ignoramous
10 hours ago
> wireguard itself has a problem that the crypto suite it uses is not supported by hardware accelerators. So if we want to get into the hundreds of gigabits range, we will possibly need to switch packet formats entirely
Has hardware-offload (for AES et al) got faster still, or that keeping CPU busy in the data path for 100gbps workloads is not ideal, or something else? The WireGuard website claims ChaPoly is at least as fast as hardware-accelerated AES. And that it can be further sped up with SIMD.
https://news.ycombinator.com/item?id=35567268).
Thanks.
> Has hardware-offload (for AES et al) got faster still, or that keeping CPU busy in the data path for 100gbps workloads is not ideal, or something else?
What is referred to by the term "hardware-offload"? For NICs:
> ConnectX NICs offload and accelerate encryption/decryption at speeds up to 400Gb/s.
* https://www.nvidia.com/en-us/networking/ethernet-adapters/
There have been MACsec implementations at 800Gb/s for several years:
* https://www.rambus.com/blogs/rambus-launches-800g-macsec-mul...
* https://semiengineering.com/the-evolution-of-ethernet-to-800...
And 1.6T/3.2T as well:
* https://www.rambus.com/security/protocol-engines/macsec-ip-3...
> The WireGuard website claims ChaPoly is at least as fast as hardware-accelerated AES.
I haven’t used WireGuard but I have easily doubled OpenSSH performance by switching back to AES.