* Can iperf work with no qos? (TCP mode)
@ 2011-06-11 17:47 Rafał Miłecki
2011-06-11 17:57 ` Michael Büsch
0 siblings, 1 reply; 3+ messages in thread
From: Rafał Miłecki @ 2011-06-11 17:47 UTC (permalink / raw)
To: b43-dev, linux-wireless
I can not get any results when trying to use iperf in TCP mode with
driver not supporting qos.
Is this expected? Does iperf fill connection with so many packets,
it's not possible to get response from the iperf receiver? Should not
Linux somehow schedule packet to allow bi-direction communication,
even on really filled connection?
--
Rafa?
^ permalink raw reply [flat|nested] 3+ messages in thread
* Can iperf work with no qos? (TCP mode)
2011-06-11 17:47 Can iperf work with no qos? (TCP mode) Rafał Miłecki
@ 2011-06-11 17:57 ` Michael Büsch
2011-06-11 18:05 ` Rafał Miłecki
0 siblings, 1 reply; 3+ messages in thread
From: Michael Büsch @ 2011-06-11 17:57 UTC (permalink / raw)
To: Rafał Miłecki; +Cc: b43-dev, linux-wireless
On Sat, 11 Jun 2011 19:47:31 +0200
Rafa? Mi?ecki <zajec5@gmail.com> wrote:
> I can not get any results when trying to use iperf in TCP mode with
> driver not supporting qos.
>
> Is this expected? Does iperf fill connection with so many packets,
> it's not possible to get response from the iperf receiver? Should not
> Linux somehow schedule packet to allow bi-direction communication,
> even on really filled connection?
iperf will only fill the best effort queue. At least that's what it did
last time when I looked into the code.
You can verify that with b43's ring statistics.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Can iperf work with no qos? (TCP mode)
2011-06-11 17:57 ` Michael Büsch
@ 2011-06-11 18:05 ` Rafał Miłecki
0 siblings, 0 replies; 3+ messages in thread
From: Rafał Miłecki @ 2011-06-11 18:05 UTC (permalink / raw)
To: Michael Büsch; +Cc: b43-dev, linux-wireless
W dniu 11 czerwca 2011 19:57 u?ytkownik Michael B?sch <m@bues.ch> napisa?:
> On Sat, 11 Jun 2011 19:47:31 +0200
> Rafa? Mi?ecki <zajec5@gmail.com> wrote:
>
>> I can not get any results when trying to use iperf in TCP mode with
>> driver not supporting qos.
>>
>> Is this expected? Does iperf fill connection with so many packets,
>> it's not possible to get response from the iperf receiver? Should not
>> Linux somehow schedule packet to allow bi-direction communication,
>> even on really filled connection?
>
> iperf will only fill the best effort queue. At least that's what it did
> last time when I looked into the code.
> You can verify that with b43's ring statistics.
It's BE queue we create for qos=0 in b43:
else
q = dev->pio.tx_queue_AC_BE;
so this should not limit iperf
As another test, I've just tried dma mode with qos=0. iperf was fine.
So now we know iperf does not need multiple queues. It seems that:
GOOD: b43 & dma & no qos
BAD: b43 & pio & no qos
In the second case connection is started (I can see that from "iperf
-s" on my receiving machine), but it seems sender does not receive any
confirmations.
--
Rafa?
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2011-06-11 18:05 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-06-11 17:47 Can iperf work with no qos? (TCP mode) Rafał Miłecki
2011-06-11 17:57 ` Michael Büsch
2011-06-11 18:05 ` Rafał Miłecki
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).