12 Apr
2017
12 Apr
'17
10:12 p.m.
Hi all
A quick question here. I’m the opposite of an expert on protocol technicalities like layer
2, layer 3, etc… but I get the general idea of LiMe having both layer 2 and 3 routing, and
such.
I’m wondering though, since every node, at least in our network, has the 10.13.0.1 IP on
the lan side, how are conflicts avoided or solved when a device with IP 10.13.0.108 - for
example - roams to a WLAN of a node where a device with such IP is already present? Or are
IP leases monitored network-wide to avoid two devices having the same IP at the same time
anywhere on the network? And if so, how big a pool do you need to always accomodate such
vast numbers of clients?
I hope this makes sense and is not based on completely wrong assumptions or
understandings.
Thank you in advance
Nk
13 Apr
13 Apr
10:08 a.m.
On 04/13/2017 12:12 AM, Nk via lime-users wrote:
I’m wondering though, since every node, at least in
our network, has the
10.13.0.1 IP on the lan side, how are conflicts avoided or solved when a
device with IP 10.13.0.108 - for example - roams to a WLAN of a node
where a device with such IP is already present? Or are IP leases
monitored network-wide to avoid two devices having the same IP at the
same time anywhere on the network?
I hope I got your point...
There's a package in lime-packages named "dnsmasq-lease-share" [1] and
included in normal LiMe images, this shares a file with the leases using
Alfred (a Batman adv utility). Have a look at these files [2].
So the leases list is shared among the whole Batman adv cloud.
If you move from a Batman cloud to another (which is less usual) the
collisions are avoided (well, reduced enough) by the %N1 parameter in
10.%N1.0.0/16 [3].
Moreover there should be some kind of check of IP availability any time
an IP is assigned, don't know really.
And if so, how big a pool do you need
to always accomodate such vast numbers of clients?
Mmh?
As there's this lease share mechanism, you don't need a pool much bigger
than the number of your clients.
In absence of such mechanism, you could calculate the probability of a
collision as in "birthday paradox" [4].
Bye!
Ilario
[1]
https://github.com/libremesh/lime-packages/tree/develop/packages/dnsmasq-le…
[2]
https://github.com/libremesh/lime-packages/blob/develop/packages/dnsmasq-le…
[3]
https://github.com/libremesh/lime-packages/blob/develop/packages/lime-syste…
[4] https://en.wikipedia.org/wiki/Birthday_problem
10:54 a.m.
Hi Ilario
Thank you so much for taking the time.
So within the same network the DHCP assignments are unique independently of what node the
client is connected to, correct? If you move between different networks, you’re saying a
conflict could of course arise. Is this package available in standard openwrt? If so could
it be used in a non-lime network where [long story short] there are a bunch of routers/APs
[with mesh nodes on the wan side] all with the same IP and subnet and DHCP range being
offered to clients [similarly to lime nodes all having 10.13.0.1 etc…], in order to have a
distributed map of active leases and prevent other routers/APs from assigning that lease
to one of their clients, or is this plugin relying on the layer 2 routing in lime?
Since you’ve been so nice to explain this, could I ask you for a followup [if you have
time] about an old thread I opened called "Speed/latency issues” from the 13th of
february? I can’t understand why lime has been performing so well for me [60mbps
speedtest.net over 2.4GHz between two TL WR1043NDs meshing with each other] only once -
literally, and terribly after that [3 to 15 mbps, 80ms latency (instead of 2ms with olsr),
etc…]. More details on the original post. Only if you have a few minutes, thank you so
much.
Nk
On 13 apr 2017, 12:08 +0200, Ilario Gelmetti <iochesonome(a)gmail.com>om>, wrote:
On 04/13/2017 12:12 AM, Nk via lime-users wrote:
I’m wondering though, since every node, at least
in our network, has the
10.13.0.1 IP on the lan side, how are conflicts avoided or solved when a
device with IP 10.13.0.108 - for example - roams to a WLAN of a node
where a device with such IP is already present? Or are IP leases
monitored network-wide to avoid two devices having the same IP at the
same time anywhere on the network?
I hope I got your point...
There's a package in lime-packages named "dnsmasq-lease-share" [1] and
included in normal LiMe images, this shares a file with the leases using
Alfred (a Batman adv utility). Have a look at these files [2].
So the leases list is shared among the whole Batman adv cloud.
If you move from a Batman cloud to another (which is less usual) the
collisions are avoided (well, reduced enough) by the %N1 parameter in
10.%N1.0.0/16 [3].
Moreover there should be some kind of check of IP availability any time
an IP is assigned, don't know really.
And if so, how big a pool do you need
to always accomodate such vast numbers of clients?
Mmh?
As there's this lease share mechanism, you don't need a pool much bigger
than the number of your clients.
In absence of such mechanism, you could calculate the probability of a
collision as in "birthday paradox" [4].
Bye!
Ilario
[1]
https://github.com/libremesh/lime-packages/tree/develop/packages/dnsmasq-le…
[2]
https://github.com/libremesh/lime-packages/blob/develop/packages/dnsmasq-le…
[3]
https://github.com/libremesh/lime-packages/blob/develop/packages/lime-syste…
[4] https://en.wikipedia.org/wiki/Birthday_problem
_______________________________________________
lime-users mailing list
lime-users(a)lists.libremesh.org
https://lists.libremesh.org/mailman/listinfo/lime-users 
12:03 p.m.
On Thursday 13 April 2017 12:54:13 Nk via lime-users wrote:
So within the same network the DHCP assignments are
unique independently of
what node the client is connected to, correct?
Yeah
If you move between
different networks, you’re saying a conflict could of course arise.
No, different network on libremesh has different broadcast domain so no IP
conflict is possible
Is this
package available in standard openwrt?
No, it is in libremesh repository
If so could it be used in a non-lime
network where [long story short] there are a bunch of routers/APs [with
mesh nodes on the wan side] all with the same IP and subnet and DHCP range
being offered to clients [similarly to lime nodes all having 10.13.0.1
etc…], in order to have a distributed map of active leases and prevent
other routers/APs from assigning that lease to one of their clients, or is
this plugin relying on the layer 2 routing in lime?
ATM it rely on alfred, that rely on batman-adv it don't work on any network,
but alfred has proved to not scale well on our network so we are considering
different mechanisms for the near future, it would be nice if we find/implement
a solution that do not rely on specific routing protocol, as this is a more
general problem that make sense to solve on any multi-DHCP L2 network
Cheers!
2810
days inactive
2811
days old
lime-users@lists.libremesh.org
3 comments
3 participants
participants (3)
-
Gio -
Ilario Gelmetti -
Nk