Re: Raspberry Pi 3B+ and pitiful network speeds

From: MJ <>
Date: Thu, 24 Jun 2021 11:14:56 UTC
On 24/06/2021 6:51 pm, Paul Zimmerman wrote:
>> On 22/06/2021 10:37 pm, Denis Ovsienko wrote:
>>> On Tue, 22 Jun 2021 08:24:02 +1000
>>> MJ <> wrote:
>>>> [  1] 0.00-30.19 sec   310 MBytes  86.0 Mbits/sec
>>>> Now, I could run these very same tests off Devuan if you like? This
>>>> may help disprove any network issues? Although, as I've stated, the
>>>> Devuan version running on the RPI3 consistently gets up to 3 times
>>>> this speed.
>>> Thank you for clarifying the test results Matt. From the numbers it
>>> indeed looks like the TCP throughput never reaches even 100Mb/s. Whilst
>>> the root cause of this might be a driver issue, or a consequence of the
>>> "early access" board USB implementation not working well in FreeBSD, it
>>> would help to eliminate trivial explanations as well.
>> Thank you too Denis for your guidance in testing this appliance.
>> The following is also addressed to all those interested:
>> Just as a follow up, I performed some tests this morning using the old Devuan SD Card and a > new NetBSD \
>> 9.2.
>> The summary is:
>> Devuan/Linux is hands down faster. NetBSD 9.2 is almost 50% faster than FreeBSD 13.
>> While NetBSD's result is not stellar it's still better than the pitiful FreeBSD throughput.
>> I will likely abandon using FreeBSD, but I should probably raise a bug report for this? I am not sure of \
>> the process for this or whether it's even wanted. That is, does FreeBSD care enough about little \
>> appliances like the Raspberry Pis to spend time fixing an obviously broken network driver/stack?
> I think the performance difference you are seeing vs. NetBSD might be
> because the FreeBSD USB driver for the RPI3 and below does not use
> interrupts, so all the USB transfers are polled. The NetBSD USB driver
> is borrowed from OpenBSD, which in turn is borrowed from the mainline
> Linux driver, which does use interrupts.
> - Paul

Thanks Paul.

In that case it seems like a waste of time posting a bug report when the driver is designed that way and only a major re-design will fix