netdev - Re: [PATCH net-next 1/1] net: usb: qmi_wwan: add default rx_urb

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAKfDRXgGw0s=DAOsR5x7SXtr6twda5U_uOEb_VNZ-0hVEEvuYg@mail.gmail.com>
Date:   Fri, 13 Nov 2020 09:50:40 +0100
From:   Kristian Evensen <kristian.evensen@...il.com>
To:     Carl Yin(殷张成) <carl.yin@...ctel.com>
Cc:     Daniele Palmas <dnlplm@...il.com>,
        Bjørn Mork <bjorn@...k.no>,
        Paul Gildea <paul.gildea@...il.com>,
        "David S . Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Network Development <netdev@...r.kernel.org>,
        linux-usb <linux-usb@...r.kernel.org>
Subject: Re: [PATCH net-next 1/1] net: usb: qmi_wwan: add default rx_urb_size

Hi Carl,

Thanks a lot for your reply.

On Fri, Nov 13, 2020 at 9:37 AM Carl Yin(殷张成) <carl.yin@...ctel.com> wrote:
>         For openwrt device, the ' Performance bottleneck ' usually is NAT, not usbnet.
>         As I remember: MT7621 have dual core, and support Hardware acceleration of 'NAT'.

Yes, you are right in that NAT can have a large effect on performance,
especially when you start being CPU-limited. However,when using perf
to profile the kernel during my tests, no function related to
netfilter/conntrack appeared very high on the list. I would also
expect the modem to at least reach the performance of the dongle, with
offloading being switched off. However, there could be some detail I
missed.

>         It seems r8152 is a pure Ethernet card, does it can use the ' Hardware acceleration '

I will do some experiments with hardware NAT, thanks for reminding me
of this feature. However, my experience with it is not very good, so I
would ideally like to find a solution that does not rely on this
feature.

>         And do you use 'mpstat -P ALL 2' to monitor each core's loading?
>         Generally USB interrupt occurs at cpu0, and the 'NAT' is also on cpu0.
>         You can try to use "echo 2 > /sys/class/net/wwan0/ /queues/rx-0/rps_cpus " to move NAT to cpu1.

I use htop to monitor the load on each core. rx_cpus is set to "e" to
balance traffic better across all cores, locking rx to one core gave a
much worse result (something like 170 Mbit/s).

>         X55 max support 31KB, there are benefit from 16KB -> 31KB.
>         Maybe your X55's FW version is old, only generates 16KB data.
>         And URB size is 32KB, but X55 only output 16KB, so maybe there are not enough number of URBs?

I had the same theory and asked my FAE if a more recent firmware is
available for my device (some of my tests were done with Quectel
RM500Q). I do not know what is the trigger for the device to generate
32KB URBs is? The fastest network I have access to gives a speed of
700-800 Mbit/s, but I do not know if that is enough?

Kristian