That are all not vald ways to measure network performance because protocols like SSH or SMB do add a lot of overhead. Install iperf3 and measure performance with that. Just start iperf3 -s
on your VF2 and on any other computer then do:
iperf3 -c <ip of the vf2>
I have done so and here are my measurements which are quite ok:
i no interessed @ banchmark test to obtain only big numbers… i need it to use in my infrastructure @ days operetionals… but if the boards give me more limitations to utilized it… @ full speed… no samba… no filezilla… i think it no interessed me… i use this tools all days… i have this infrastructure… and this board must be integrated into this enviroment… if this integration not possible… i thik use normale motherboard with X86_64 tech.
last one… normal share by 2 hosts Win10 cabled… into my lan… shared files over 80M/s… simple, fast,
If the semplicity is losted… the project not have sanse…
… visionfive is a fantastic board pheraps… i not need it…
Well, explanation has already been given by @cwt
It is totally ok, if only a specific application is of interest to you. But I do wonder why you did then buy a development board using a new ISA in the first place.
And it is clearly advertised as a such: https://www.starfivetech.com/en/site/boards
make VisionFive 2 the best affordable RISC-V development board
You’re right about that. This board isn’t intended as a replacement for any x86 mini-PC or home server, not even close to the Raspberry Pi. It has limitations in both hardware and software. However, the purpose of this board is clearly for RISC-V development. If you try building a program or rebuilding the Linux kernel on this board (with NVMe), you’ll find its performance is surprisingly good, and memory usage during the build process is also efficient.
mmmm i suppose you have a problem on your atom-board… today… 2 host with celeron J1900… and debian 11 run to 82M/s… with simple samba condivision…
in the future i return on this project…hoping… in a big step…
2years ago… i used a ASUS eeebox with Intel N450 or N455 i don’t remeber it, and gigabit ethernet… and a old Debian 9… same result had today… 80M/s… but is died… 365 24/24 after 4 years… bye bye
…“Now my rsync is 2.5x faster, but still slower than 100Mbit/s”…
i share my files over 11Megabyte per second whit Visionfive2 (for cleareance… stable into 11,8-11,9 Megabyte per second)… i sature all 100Mbit/s bandiwth… and you minus than me?? with this trick/configuration?! ok…
thankyou
As I said, the bottleneck is the OpenSSL that cannot utilized the hardware crypto acceleration. And that was 8 months ago! Things changed a lot on both kernel and userspace.
If you do not warry about security , you can try old openssl with rc4(named acfour) .
And very old openssl have none-crypt will give you fastest connection. But I think It will not be built on “morden” linux.
The above line should transfers 42 files at a time in parallel over 42 ssh connections which will use multiple CPU’s and reuse the existing ssh connections instead of wasting resources destroying and creating a new connection for each and every file. You can just bump up or down the random number I selected to tune the performance until either your source/destination CPU’s are overloaded or your network link is saturated One of the three will be the bottleneck (source CPU/destination CPU/network link between machines).
In my opinion vf2 is really only for riscv enthusiast. The official software stack are still hacky at this point, and in some cases, non-performant. The kernel is based on 5.15.0, which doesn’t have a patch for memcpy/memmove in kernel, and is using unaligned memory access, which hurt performance a lot. Also there’s no vector extension in jh7110/vf2. I bought it just because I want to explore riscv, and a rpi3-ish speed is good enough for me.
Back to your question: make sure you use the latest release VisionFive 2 Debian Wiki - 202311 Release | RVspace flash it to an sdcard and boot from it. In this way you’ll be using U-boot/OpenSBI/kernel all from the latest release. There are code within u-boot to configure gmac and phy to different speed based on board revision, and there’s also code in u-boot to patch the device-tree, which gets passed to Linux kernel, and does some gmac quirk based on board revision.