[CUWiN-Dev] notes on TDV

[Matt Isaacs]

Routes to the ethernet interfaces on neighboring nodes are showing up in the
routing tables onthe gateway nodes.  Is this the correct behavior?  My
initial reaction is no.

I don't think its causing any problems, it just surprised me.

On 10/30/06, Matt Isaacs <isaacsm at cuwireless.net> wrote:
>
> Turning off IP6 in HSLS has cut loads almost in half.
>
> On 10/30/06, Matt Isaacs <isaacsm at cuwireless.net> wrote:
> >
> > Here's what I've been running to test the node's connections:
> >
> > netstat -rn -f inet | awk '/^169\.254\./ { print $1; }' | xargs -n 1
> > ping -c 100 -n | tee pings.txt
> >
> > I then use this to extract the parts I'm interested in:
> >
> > cat pings.txt | grep "Statistics\|packets\|round-trip" > sum.txt
> >
> > If anyone sees a problem with this or knows of a better way, please let
> > me know.
> >
> > I don't ping the 10.0 addresses as its impossible to get an accurate
> > measure due to route fluctations.
> >
> > On 10/30/06, Matt Isaacs <isaacsm at cuwireless.net> wrote:
> > >
> > > The tpscale isn't going to work as well as we had hoped, at least not
> > > at 3.  The node appears to see nothing but itself at 3.  If we slide up to a
> > > scale of 2, we see more stuff but poorly.  We get ~75& packet failure rate
> > > to nodes that are directly next to it.  (I've been testing from a gateway.)
> > > I'm trying it now with 1.  What I've seen suggests ~50% packet failure rate
> > > to neighboring nodes.  No scaling yeilds about a 30% failure rate to
> > > neighboring nodes.
> > >
> > > I'll try turning off IP6 and see what that does for us.
> > >
> > > On 10/30/06, David Young < dyoung at pobox.com> wrote:
> > > >
> > > > On Sun, Oct 29, 2006 at 09:38:39PM -0600, David Young wrote:
> > > > > Some routers at TDV are spending about 50% of their CPU time in
> > > > the
> > > > > operating system, leaving very little CPU headroom for forwarding,
> > > > HSLS,
> > > > > etc.  I believe we owe the system overhead to the density of the
> > > > network;
> > > > > every node can hear every other, so every node is bombarded by
> > > > HSLS
> > > > > packets.  There are a few things we can try to get the CPU usage
> > > > down:
> > > > >
> > > > >         1 Turn off IPv6 (easily halves the number of packets).
> > > >
> > > > Turned off in the TDV build.
> > > >
> > > > >         2 Scale transmit power (sysctl -w hw.ath0.tpscale=4) to
> > > > control
> > > > >           nodal degree and interference.
> > > >
> > > > I set tpscale to 3 on the TDV build.  I hope that isn't too much.
> > > >
> > > > >         3 Figure out how to adjust receive sensitivity to filter
> > > > >           out transmissions from distant nodes.
> > > >
> > > > Looks to me like the HAL does not provide an API for that.
> > > >
> > > > > Matt and I noticed that the diversity sysctl is turning back on
> > > > > (hw.ath0.diversity = 0 -> 1) when the ath0 is brought up.  I.e.,
> > > > we
> > > > > need to put '!/sbin/sysctl -w hw.$int.diversity=0' at the bottom
> > > > of
> > > > > ifconfig.backhaul.tmpl.  I don't know if this is making any
> > > > practical
> > > > > difference, since we have forced the interfaces to use just one
> > > > antenna
> > > > > for transmit and receive, and those settings are not reset when we
> > > > bring
> > > > > the interface up.
> > > >
> > > > Done.
> > > >
> > > > > Matt tells me that dhcpselect is not working right on
> > > > gateways.  If the
> > > > > DHCP server is not available when the node boots, dhcpselect may
> > > > or
> > > > > may not detect an Internet connection when the DHCP server starts.
> > > > > Sounds like a regression in dhcpselect.
> > > >
> > > > I haven't investigated this, yet.
> > > >
> > > > Dave
> > > >
> > > > --
> > > > David Young             OJC Technologies
> > > > dyoung at ojctech.com      Urbana, IL * (217) 278-3933
> > > > _______________________________________________
> > > > CU-Wireless-Dev mailing list
> > > > CU-Wireless-Dev at lists.cuwireless.net
> > > > http://lists.chambana.net/cgi-bin/listinfo/cu-wireless-dev
> > > >
> > >
> > >
> > >
> > > --
> > > Matthew Isaacs
> > >
> > > CUWiN Network Engineer
> >
> >
> >
> >
> > --
> > Matthew Isaacs
> >
> > CUWiN Network Engineer
> >
>
>
>
> --
> Matthew Isaacs
>
> CUWiN Network Engineer
>



-- 
Matthew Isaacs

CUWiN Network Engineer
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.chambana.net/mailman/archive/cu-wireless-dev/attachments/20061030/5329d525/attachment.html