netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* fq_codel with 802.15.4 6LoWPAN fragmentation issue
@ 2015-01-16 12:02 Alexander Aring
  2015-01-16 12:57 ` Eric Dumazet
  0 siblings, 1 reply; 4+ messages in thread
From: Alexander Aring @ 2015-01-16 12:02 UTC (permalink / raw)
  To: netdev; +Cc: linux-wpan, stephen, jukka.rissanen

Hi,

when using "fq_codel" as network scheduler, I noticed an issue with IEEE
802.15.4 6LoWPAN fragmentation. The issue is that high payloads IPv6
packets are dropped.

I detected the issue on the sending side which segments the 6LoWPAN
packet into several fragments.

In the IEEE 802.15.4 6LoWPAN fragmentation we have a 6LoWPAN interface
(MTU 1280) on top of an IEEE 802.15.4 interface (MTU 127). The 6LoWPAN
fragmentation algorithm creates in the "ndo_start_xmit" netdev_ops
callback function of the 6LoWPAN interface new sk_buffs for the
underlaying IEEE 802.15.4 interface (as skb->dev) and sends them via a
dev_queue_xmit.

IEEE 802.15.4 is a very slow connection and after each transmit there is
an "interframe spacing time" which depends on frequency setting. In my
testcase, that's about ~500 us for a payload above 18 bytes. This is
done by calling netif_stop_queue for the interface and start a timer
after each transmitted frame according to the interframe spacing time.
When the timer fires, it wakes the queue again by calling
netif_wake_queue.

Now: What I am noticed with fq_codel is that the code at [0] will drop
high payloaded IPv6 packets. Of course, if we have an IPv6 packet of
more than 1280 bytes payload we have a fragmentation (IPv6
fragmentation) over a fragmentation (802.15.4 6LoWPAN fragmentation).

When I increase the "params->interval" parameter I can send a "longer"
IPv6 packet depending how much I increase the "params->interval"
parameter.

I assume that the "params->interval" which defaults to MS2TIME(100) is
too small. I further assume that at the beginning of fragmentation the
interval measurement starts "vars->first_above_time = now +
params->interval;" (line 247 at [0]) and while sending several fragments
with the above mentioned interframe spacing time, the sending takes too
long and the skb will be dropped.

My questions are:

- What's the best way to deal with something like that?

- Is there a way to set the "params->interval" _per interface_? So I can
  change the default parameter to a suitable value in the "ndo_init"
  callback of the 6LoWPAN interface.

Thanks in advance.

- Alex

[0] http://lxr.free-electrons.com/source/include/net/codel.h#L248

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: fq_codel with 802.15.4 6LoWPAN fragmentation issue
  2015-01-16 12:02 fq_codel with 802.15.4 6LoWPAN fragmentation issue Alexander Aring
@ 2015-01-16 12:57 ` Eric Dumazet
  2015-01-16 13:00   ` Eric Dumazet
  0 siblings, 1 reply; 4+ messages in thread
From: Eric Dumazet @ 2015-01-16 12:57 UTC (permalink / raw)
  To: Alexander Aring; +Cc: netdev, linux-wpan, stephen, jukka.rissanen

On Fri, 2015-01-16 at 13:02 +0100, Alexander Aring wrote:
> Hi,
> 
> when using "fq_codel" as network scheduler, I noticed an issue with IEEE
> 802.15.4 6LoWPAN fragmentation. The issue is that high payloads IPv6
> packets are dropped.
> 
> I detected the issue on the sending side which segments the 6LoWPAN
> packet into several fragments.
> 
> In the IEEE 802.15.4 6LoWPAN fragmentation we have a 6LoWPAN interface
> (MTU 1280) on top of an IEEE 802.15.4 interface (MTU 127). The 6LoWPAN
> fragmentation algorithm creates in the "ndo_start_xmit" netdev_ops
> callback function of the 6LoWPAN interface new sk_buffs for the
> underlaying IEEE 802.15.4 interface (as skb->dev) and sends them via a
> dev_queue_xmit.
> 
> IEEE 802.15.4 is a very slow connection and after each transmit there is
> an "interframe spacing time" which depends on frequency setting. In my
> testcase, that's about ~500 us for a payload above 18 bytes. This is
> done by calling netif_stop_queue for the interface and start a timer
> after each transmitted frame according to the interframe spacing time.
> When the timer fires, it wakes the queue again by calling
> netif_wake_queue.
> 
> Now: What I am noticed with fq_codel is that the code at [0] will drop
> high payloaded IPv6 packets. Of course, if we have an IPv6 packet of
> more than 1280 bytes payload we have a fragmentation (IPv6
> fragmentation) over a fragmentation (802.15.4 6LoWPAN fragmentation).
> 
> When I increase the "params->interval" parameter I can send a "longer"
> IPv6 packet depending how much I increase the "params->interval"
> parameter.
> 
> I assume that the "params->interval" which defaults to MS2TIME(100) is
> too small. I further assume that at the beginning of fragmentation the
> interval measurement starts "vars->first_above_time = now +
> params->interval;" (line 247 at [0]) and while sending several fragments
> with the above mentioned interframe spacing time, the sending takes too
> long and the skb will be dropped.
> 
> My questions are:
> 
> - What's the best way to deal with something like that?
> 
> - Is there a way to set the "params->interval" _per interface_? So I can
>   change the default parameter to a suitable value in the "ndo_init"
>   callback of the 6LoWPAN interface.
> 
> Thanks in advance.

Sure, fq_codel is a qdisc, each network device gets its own qdisc
hierarchy.

You can therefore set different parameter on your interface,

Very simple example :

tc qdisc dev lowpan0 root fq_codel interval XXX

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: fq_codel with 802.15.4 6LoWPAN fragmentation issue
  2015-01-16 12:57 ` Eric Dumazet
@ 2015-01-16 13:00   ` Eric Dumazet
  2015-01-16 13:04     ` Eric Dumazet
  0 siblings, 1 reply; 4+ messages in thread
From: Eric Dumazet @ 2015-01-16 13:00 UTC (permalink / raw)
  To: Alexander Aring; +Cc: netdev, linux-wpan, stephen, jukka.rissanen

On Fri, 2015-01-16 at 04:57 -0800, Eric Dumazet wrote:
>  example :
> 

Sorry, one keyword was missing :

tc qdisc replace dev lowpan0 root fq_codel interval XXX

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: fq_codel with 802.15.4 6LoWPAN fragmentation issue
  2015-01-16 13:00   ` Eric Dumazet
@ 2015-01-16 13:04     ` Eric Dumazet
  0 siblings, 0 replies; 4+ messages in thread
From: Eric Dumazet @ 2015-01-16 13:04 UTC (permalink / raw)
  To: Alexander Aring; +Cc: netdev, linux-wpan, stephen, jukka.rissanen

On Fri, 2015-01-16 at 05:00 -0800, Eric Dumazet wrote:
> On Fri, 2015-01-16 at 04:57 -0800, Eric Dumazet wrote:
> >  example :
> > 
> 
> Sorry, one keyword was missing :
> 
> tc qdisc replace dev lowpan0 root fq_codel interval XXX
> 

But the parameter that you probably want to change is not the interval.
It is probably the target :

tc qdisc replace dev lowpan0 root fq_codel interval 200ms target 25ms

Now, if the fragmentation was done after qdisc dequeue, as in GSO layer,
you would not have this issue : all segments of one ipv6 message would
be transmitted, no matter what.

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2015-01-16 13:04 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-01-16 12:02 fq_codel with 802.15.4 6LoWPAN fragmentation issue Alexander Aring
2015-01-16 12:57 ` Eric Dumazet
2015-01-16 13:00   ` Eric Dumazet
2015-01-16 13:04     ` Eric Dumazet

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).