Re: [PATCH v2] ath10k: Implement get_expected_throughput callback

On 2018-03-26 12:52, Sven Eckelmann wrote:

On Freitag, 23. März 2018 19:37:14 CEST Anilkumar Kolli wrote:

quoted

+static u32 ath10k_get_expected_throughput(struct ieee80211_hw *hw,
+                                         struct ieee80211_sta *sta)
+{
+       struct ath10k_sta *arsta = (struct ath10k_sta *)sta->drv_priv;
+
+       return ewma_sta_txrate_read(&arsta->ave_sta_txrate);
+}

On Freitag, 23. März 2018 19:11:48 CEST akolli@codeaurora.org wrote:

quoted

quoted

Antonio and Felix, please correct me when this statement is incorrect.

The expected_throughput as initially implemented for minstrel(_ht) is
not
about the raw physical bitrate but about the throughput which is
expected for
things running on top of the wifi link. See
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=cca674d47e59665630f3005291b61bb883015fc5
for more details

when I interpret your change correctly then your it doesn't get the
information about packet loss or aggregation and doesn't do anything
convert
from raw physical rate to something the user could get see. It will
just
overestimate the throughput for ath10k links and thus give wrong
information
to routing algorithms. This could for example cause them to prefer
links over
ath10k based hw when mt76 would actually provide a significant better
throughput.

Beside that - why is the ave_sta_txrate only filled when with new
information
when someone requests the current expected_throughput via
get_expected_throughput. I would have expected that it is filled
everytime you
get new information about the current rate from the firmware
(ath10k_sta_statistics).

Yes. ideally it should be doing the rate avg. of all the sent packets.

No, not the PHY rate average - but the "throughput avg". And the "ideally"
here sounds a little bit like in "Our medical doctor would ideally not
decapitate each patient but we have at least an MD".

The rate average and throughput are relative. no?

- Anil.