[Bloat] CPU consumption using TC-TBF and TC-POLICE to limit rate

[Y]

Hello

I heard that IBF costs cpu load.

How about shaping only egress with TBF?

Yutaka.

On 2020/05/26 18:47, Jose Blanquicet wrote:
> Hi everyone,
>
> We have an embedded system with limited CPU resources that acts as
> gateway to provide Internet access from LTE to a private Wi-Fi
> network. Our problem is that the bandwidth on LTE and Wi-Fi links is
> higher than what the system is able to handle thus it reaches 100% of
> CPU load when we perform a simple speed test from a device connected
> to our Wi-Fi Hotspot.
>
> Therefore, we want to limit the bandwidth to avoid system gets
> saturated is such use-case. To do so, we thought to use the QDISC-TBF
> on the Wi-Fi interface. For instance, to have 10Mbps:
>
>      tc qdisc add dev wlan0 root tbf rate 10mbit burst 12500b latency 50ms
>
> It worked correctly and maximum rate was limited to 10Mbps. However,
> we noticed that the CPU load added by the TBF was not negligible for
> our system. In fact, we compared the CPU load when limitation was done
> by TBF and by the device on the private network, e.g. wget tool with
> parameter "--limit-rate". As result, we found that the CPU load when
> using TBF was 10-15% higher.
>
> Then, we thought that using traffic shaping in egress, packets need to
> be un-natted (which takes CPU) and pass through the system to then get
> dropped. Therefore, we tried to use an incoming policer instead of
> egress traffic shaping as following:
>
>      tc qdisc add dev eth0 ingress handle ffff:
>      tc filter add dev eth0 parent ffff: protocol ip u32 match u32 0 0
> police rate 10mbit burst 1m drop
>
> Unfortunately, as per egress traffic shaping, we still obtained a high
> CPU cost because of rate limiting. However, also in this case, we are
> not sure we chose the most efficient option in terms of CPU cost to
> police in ingress.
>
> Given that, we were wondering if we are doing wrong by choosing TBF?
> Or maybe we are using wrong parameters? We found everywhere that TBF
> is the simplest way to limit the rate thus we suppose it is also the
> most efficient QDISC. Is our supposition correct? Or there is another
> way to limit rate well-known by its low CPU consumption? Any
> suggestion is welcome, just taking into account we are using
> libnl-3.2.28 and linux-kernel 3.18. In case, we could change libnl but
> not kernel version, at most some specific patches.
>
> Thanks in advance for the support!
>
> Jose Blanquicet
> _______________________________________________
> Bloat mailing list
> Bloat at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/bloat/attachments/20200526/a760f2f7/attachment.html>