Re: [B.A.T.M.A.N.] 11ac throughput on mesh links

[Sven Eckelmann <sven@narfation.org>]

[-- Attachment #1: Type: text/plain, Size: 2107 bytes --]

On Montag, 1. Januar 2018 13:12:53 CET Marek Lindner wrote:
[...]
> This is a known bug caused by the QCA WiFi driver firmware blob. The exported
> TX bitrate value is utterly bogus. Only QCA is in the position to fix that.
> 
> There have been attempts such as this one:
>  https://github.com/torvalds/linux/commit/c1dd8016ae02557e4f3fcf7339865924d9357f76
> Not sure this fix addresses your case. Sven might know.

This only works for 10.4 firmware versions with peer stats enabled. The 10.2.4 
firmware versions (only some are actually supported) require following patchset:

* https://patchwork.kernel.org/patch/10092915/
* https://patchwork.kernel.org/patch/10092917/
* https://patchwork.kernel.org/patch/10092919/

And you need the patch mentioned by Marek (+ the patch referenced in it) to 
get any TX rate values at all.

But QCA already knows that they (relatively often) still report completely 
bogus values (and only QCA can fix it). And you must understand that the 
values which you will get here are *a lot* higher than what you can 
realistically achieve via this link since the TX/RX data rates are actually 
physical data rates and not the throughput measured via TCP/UDP/... - or by
looking at the payload of the actually transported (QoS) data packets.

So let as assume for now that you will lose 50% of the physical data rate to 
some expected overhead. Then you might still have the problem that each packet 
has to be retransmitted 4 times (or more) before it can be received by the 
other end and the aggregation is 1. The TX/RX data rate information in iw will 
not capture anything of that.

But there can also be a lot of other factors which influence the performance - 
maybe you CPU is not capable of handling the packet generation and 
transmission, the MTU is not configured properly on the slave device, the NIC 
might not be able to transmit a single flow to saturate the link, batman-adv's 
throughput meter might not handle some packet loss as well as expected, the 
qdisc (flow dissector) might fail to handle the batman-adv tp packets 
properly, ...

Kind regards,
	Sven

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]