I'm finding that ssh on my Gigabyte MP30-AR1 is rather slow, presumably because there are no AES instructions or mitigating optimisations for AES. Is NEON AES implemented in OpenSSL / OpenSSH package builds on EL8 on aarch64?
All of the old ciphers that were faster (e.g. arcfour) have long been removed from OpenSSH, so does anyone have any other tricks for getting more out of ssh on aarch64? I'm specifically looking for an ssh optimisation here, I'm hoping to avoid implementing a bypass hack in the workflow (e.g. setting up netcat over ssh and using that for transfers).
Many thanks.
Gordan
https://linuxaria.com/article/tar-rsync-netcat-scp has an interesting discussion about the speed of different file transfer methods. It also has some comments where suggestions about settings and options to increase speed are offered.
Not sure what your use case is but perhaps this will help.
Ron
On 2020-07-01 1:50 p.m., Gordan Bobic wrote:
I'm finding that ssh on my Gigabyte MP30-AR1 is rather slow, presumably because there are no AES instructions or mitigating optimisations for AES. Is NEON AES implemented in OpenSSL / OpenSSH package builds on EL8 on aarch64?
All of the old ciphers that were faster (e.g. arcfour) have long been removed from OpenSSH, so does anyone have any other tricks for getting more out of ssh on aarch64? I'm specifically looking for an ssh optimisation here, I'm hoping to avoid implementing a bypass hack in the workflow (e.g. setting up netcat over ssh and using that for transfers).
Many thanks.
Gordan
Arm-dev mailing list Arm-dev@centos.org https://lists.centos.org/mailman/listinfo/arm-dev
I am aware of that, but there have been NEON AES optimisations floating around, e.g.: https://www.linaro.org/blog/accelerated-aes-for-the-arm64-linux-kernel/
I guess what I'm asking is if EL8 aarch64 ships with any such thing enabled, and any other tweaks I could try to squeeze more ssh speed out of it. I'm finding that piping data over aes-128-cbc (which seems to be the fastest implemented cipher) over gigabit ethernet tops out at about 30MB/s per CPU core, with each ssh session being limited to this (even though I can run 8 sessions in parallel at that speed).
Gordan
On Wed, Jul 1, 2020 at 7:13 PM Marcin Juszkiewicz < marcin.juszkiewicz@linaro.org> wrote:
W dniu 01.07.2020 o 19:50, Gordan Bobic pisze:
I'm finding that ssh on my Gigabyte MP30-AR1 is rather slow, presumably because there are no AES instructions or mitigating optimisations for
AES.
X-Gene cpu present in your machine lacks AES instructions support. _______________________________________________ Arm-dev mailing list Arm-dev@centos.org https://lists.centos.org/mailman/listinfo/arm-dev
-
Gordan Bobic -
Marcin Juszkiewicz -
Ron Wheeler