Discussion:
Routing problem with bad ETX
Simon Ebnicher
2013-04-30 15:06:11 UTC
Permalink
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3

The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.

Can anyone confirm or explain this behaviour and maybe offer a solution?

Best regards,
Simon Ebnicher
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Michel Blais
2013-04-30 15:21:53 UTC
Permalink
That normal since 2 up without lost will give you a cost of 2. The
single up link cost can be tweeked via LinkQualityMult. If you want to
double de link cost for this link, change LinkQualityMult, for remote ip
address or default for all the subnet, to 0.5, to triple it change it to
0.33, etc.

Michel
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Saverio Proto
2013-04-30 16:04:32 UTC
Permalink
You need to adjust your multicast rate.
Btw 0.6.3 olsrd is very old. Please use this feed to compile latest version:

https://github.com/openwrt-routing/packages

best regards

Saverio Proto
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
https://lists.olsr.org/mailman/listinfo/olsr-users
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Ferry Huberts
2013-04-30 16:08:03 UTC
Permalink
Post by Saverio Proto
You need to adjust your multicast rate.
https://github.com/openwrt-routing/packages
Saverio, can you upgrade this feed to 0.6.5.3?
Thanks!
Post by Saverio Proto
best regards
Saverio Proto
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
https://lists.olsr.org/mailman/listinfo/olsr-users
--
Ferry Huberts
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Ferry Huberts
2013-04-30 16:09:28 UTC
Permalink
There are some relevant changes in there :-)

0.6.5.3 -------------------------------------------------------------------

Ferry Huberts (5):
pud: detect the java include directory
pud: better detection of java jdk
gateway: work around kernel IPIP module initialisation bug
main: use /dev/urandom by default
pud: set local loopback for multicast tx

Henning Rogge (2):
Update version after release of v0.6.5.2
Release v0.6.5.3

Ronald in 't Velt (1):
Fix setsockopt for setting Traffic Class in IPv6
Post by Ferry Huberts
Post by Saverio Proto
You need to adjust your multicast rate.
https://github.com/openwrt-routing/packages
Saverio, can you upgrade this feed to 0.6.5.3?
Thanks!
Post by Saverio Proto
best regards
Saverio Proto
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
https://lists.olsr.org/mailman/listinfo/olsr-users
--
Ferry Huberts
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Saverio Proto
2013-05-01 00:06:40 UTC
Permalink
I did it now ! :)

Saverio
Post by Ferry Huberts
Post by Saverio Proto
You need to adjust your multicast rate.
https://github.com/openwrt-routing/packages
Saverio, can you upgrade this feed to 0.6.5.3?
Thanks!
Post by Saverio Proto
best regards
Saverio Proto
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
https://lists.olsr.org/mailman/listinfo/olsr-users
--
Ferry Huberts
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Simon Ebnicher
2013-05-02 12:42:54 UTC
Permalink
Thank you very much for the push in the right direction.
We updated olsrd to 0.6.5.3 (which didn't, of course, solve the problem).
As you suggested, the problem could be solved by modifying the multicast
rate. We didn't know this was something you could change in OpenWrt.
The best tradeoff for our setup was a multicast rate of 6 MBit/s.
5.5 MBit/s still propagates too well and we suspect that the reason is
the different modulation scheme (direct sequence spread spectrum) of
that bitrate.

Best regards,
Simon Ebnicher
Post by Saverio Proto
You need to adjust your multicast rate.
https://github.com/openwrt-routing/packages
best regards
Saverio Proto
Post by Simon Ebnicher
Hi,
we have set up a small OLSR mesh network with three routers (TP-Link
TL-WDR3600 running OpenWrt 12.09rc1, olsrd 0.6.3). We placed two routers
near to each other, and the third router at quite some distance.
R1 <----ETX:~1----> R2 <---------------ETX: 1.x to 2.x-----------> R3
The ETX on the direct link between R1 and R3 is usually below 2 (about
1.3 to 1.7), so the direct route is preferred. However, the direct route
is completely unusable for data transfer (e.g., ping reports >95% packet
loss). As soon as two hop route via R2 is used, ping and other traffic
works fine.
We are wondering why the unreliable link is rated too optimistically. We
suspect that the Hello and TC multicast packets (which are used to
calculate the ETX) may have a better reception probability than unicast
packets (e.g., because they are sent at lower rates, no RTS/CTS). This
would lead to ETX values that do not reflect the link accurately.
Can anyone confirm or explain this behaviour and maybe offer a solution?
Best regards,
Simon Ebnicher
--
Olsr-users mailing list
https://lists.olsr.org/mailman/listinfo/olsr-users
--
Olsr-users mailing list
Olsr-***@lists.olsr.org
https://lists.olsr.org/mailman/listinfo/olsr-users
Loading...