[ath9k-devel] First frame losses in A-MPDUs

[ath9k-devel] First frame losses in A-MPDUs

[Andrew Heard]

Hi,

A colleague and I have noticed that the first frame in an A-MPDU is lost more frequently than 
the remaining frames. For example, I used a fixed rate of 95 (MCS 7, SGI, 40 MHz, 3 Streams)
and ran iperf between two nodes. I see the following error rates for the frame indices in an A-MPDU,
maximum of 32 frames:

0	0.209301
1	0.0741858
2	0.0725545
3	0.076915
4	0.0705244
5	0.0706704
6	0.0830724
7	0.0875648
8	0.0832975
9	0.0814545
10	0.0908066
11	0.0816703
12	0.0803201
13	0.0863222
14	0.0845363
15	0.0850065
16	0.0886615
17	0.088502
18	0.0865359
19	0.088343
20	0.0872486
21	0.0849819
22	0.0955315
23	0.0890701
24	0.0878059
25	0.0895934
26	0.0880374
27	0.0892722
28	0.091761
29	0.100914
30	0.100792
31	0.0966546

The trend we expect to see is an increase in error rate from 0 to 31, based on this paper:
http://www.mwnl.snu.ac.kr/~schoi/publication/Conferences/14-CONEXT-BYEON.pdf <http://www.mwnl.snu.ac.kr/~schoi/publication/Conferences/14-CONEXT-BYEON.pdf>
This seems to be the case, apart from the first frame.

Is anyone else experiencing similar loss patterns and/or have a possible explanation for this?

It seems to degrade throughput fairly significantly since the BlockAck window can?t be advanced
and the lost frame is software retried as an individual frame.

Thanks,

Andrew
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20151116/d8ea19be/attachment.html