On Mon, May 04, 2026 at 01:07:25PM -0400, Demi Marie Obenour wrote: > On 5/4/26 07:28, Richard Kettlewell wrote: > > On 02/05/2026 23:32, Demi Marie Obenour wrote: > >> Can you provide benchmarks showing that the accelerator is faster > >> than the CPU on realistic workloads? > > > > The consistent improvements in latency and throughput started around the > > 2Kbyte block size (10-20%) and improved as blocks grew, with around 50% > > latency reduction and 150% throughput increase at 250Kbyte blocks (which > > is close to our message size limit). > > > > Obviously this reflects the hardware we are using (which has no AES > > support in the application cores), outcomes may differ on other targets. > > Can you use ChaCha20-Poly1305 or Adiantum instead of AES? That should > be significantly faster on the CPU
Let's please wind down this sub-thread. I think it is immaterial to the original discussion whether alternative ciphers are an option in a given case. In some cases they will be, in others not. Alexander
