Re: Experiencing very slow browsing at times (atheros wireless)

[Sten Daniel Soersdal <netslists@xxxxxxxxx>]

Sepherosa Ziehau wrote:
> On 5/26/07, Sten Daniel Soersdal <netslists@gmail.com> wrote:
> > Petr Janda wrote:
> > > Sepherosa Ziehau wrote:
> > >>
> > >> Haha, the AP is broken.  Its does not use single TX seq (as required
> > >> by 802.11-1999 R2003, same as 802.11e for non-Qos data and bcast mgt)
> > >> for data and mgt packets.  Your AP seems to use one TX seq for beacon
> > >> (99% of mgt packets) and another TX seq for data.  That also explains
> > >> why using lower rate will have better result: number of retries is
> > >> reduced and duplication detection is performed only on retried
> > >> packets.
> > >>
> > >> Please test following patch and see whether it makes the situation
> > >> better in 11g mode:
> > >> http://leaf.dragonflybsd.org/~sephe/skip_beacon.diff
> > >>
> > >>
> > > Testing now, will let you know of success or failure within the next 
> > day
> > > or two. If this worked out to be the fix, I shall call you Sir Genius
> > > from now on :D
> > >
> > > Thanks!!!
> > > Petr
> > >
> >
> > Hopefully that will fix it. I'm a little puzzled by the tcp
> > conversations here. The remote server advertises an mss of 1420 (which
> > indicates an MTU of 1440) but the packets you receive is of size 1452
>
> Mmm, I think the MTU advertised by the remote server is 1460 (mss +
> ip_hdr + tcp_hdr)

yes
I forgot to mention that i was referring to the conversation between 
192.168.1.50 <-> 140.211.166.170 found in "dumpradio.pcap".

> > (which translates to mss 1432) indicating that mss is adjusted along the
>
> mss indicates that advertiser wants to receive in that size.  It is
> used to limit the segment size of other side.  The mss advertised by
> us is larger than server's mss, then I think whether the server want
> to contraint the segment size to its own mss depends on server's TCP
> implementation.

Adjusting mss along the path is also common practice among internet 
service providers who need to reduce path-mtu, to reduce or avoid 
fragmentation. Unfortunately sometimes the provider will unconditionally 
reset mss and not just adjust it to a lower value when it's too high.
Which is why i mentioned it.

--
Sten Daniel Soersdal