Hello !
I´ve been "collecting" and reading all mails and articles on equal cost multipath routing for about two month, because I have the possibility to connect to two ISPs over two different devices and want to do some primitive (per-flow based) loadbalancing. But it doesn´t work, or at least I can´t get it working. So it would be really nice if someone could explain me ho to set up ecmp from the scratch.
I am familiar with standard network and routing
issues (kernel <2.4.x), masquerading (although I´ve got static IPs),
iptables and the new iproute (ip) commands, and know how to use
traffic control and sheduling (at least in theory :-) ). I know _nothing_
about RIP, OSPF & routing daemons except their name, use and some principles
of work.
My setup is as follows:
*ISP1* *ISP2*
| |
(ppp0) +---*LinuxBox*--+ (ppp1)
|
| (eth0)
|
-----+-----+----+----+---- (LAN)
. . . .
LinuxBox is connected ...
through ppp0 to ISP1, address
10.0.1.10,
gw 10.0.1.1
(1st possible default route)
through ppp1 to
ISP2, address 10.0.2.10,
gw 10.0.2.1
(2nd possible default route)
through eth0 to LAN, address
10.0.0.1,
netmask
255.255.255.0
With this setup everything works fine. Both
systems know all of my IPs and have routing set correct.
Then I tried to set it (ecmp) up like in the examples found on the net,
but they didn´t work, and the error was always ths same: When I had
finished standard routing and finally used the "ip route add default nexthop ...
nexthop ..." command, he accepted it first, but marked one device "DEAD"
immediately.
All necessary options have been compiled into the
kernel (now 2.4.13).
I don´t know what to do and where to look any
more - theres no documentation or howto on details of ecmp. It would be really a
great help if someone knows something about it.
-=> Sebastian Bleikamp
-=> EMail: <Sebastian.Bleikamp@xxxxxxxxxxxxxxxxxxx>
-=> Phone: +49-172-6545394
-=> PGP key avaible on public key servers.
-=> EMail: <Sebastian.Bleikamp@xxxxxxxxxxxxxxxxxxx>
-=> Phone: +49-172-6545394
-=> PGP key avaible on public key servers.