Hi,
First, here is my test configuration.
I use two ER-8 (version 1.8.0) with a symetrical conf :
> Eth0 is dedicated to LAN
> Eth1 is dedicated to WAN1
> Eth2 is dedicated to WAN2
I use 2 routers to simulate WAN links.
All interfaces have static IP addresses.
I want to get a simple symetrical load-balancing between eth1 and eth 2 with 50% balance. "Simple" means without Policy Routing. "Symetrical" means that the load-balancing works on the both ER-8s.
I personalised the route-test by ping-watching the opposite interface (ER1 eth1 watchs ER2 eth1, ER2 eth1 watchs ER1 eth1, ER1 eth2 watchs ER2 eth2, ER2 eth2 watchs ER1 eth2).
To be more precise, hereby is the schematic of the test's plan.
And here attached are the two confs of the ER-8s.
For testing, I use 2 extra PCs on eth0 of each ER-8 that ping each other.
I met 3 issues :
> Pb1 : When the load-balancing is operational and the 2 WAN links are up, I noticed that the ping flow between the 2 PCs always uses the same path, and so the same WAN link. I checked it with sniffer on both central routers. So for this ping session, the 50/50% load-balancing is not effective.
Nota : Sometimes the requests go through WAN1 and the replies through WAN2.
> Pb2 : When I watch load-balancing statistics ("show load-balance status"), the "WAN In" and "WAN Out" indicators are not incremented in relation with the real number of pings between the 2 PCs. These indicators are very low. "Local Out" seems more relevant with the ping check sequence of ER-8s.
> Pb3 : The 3st issue concerns the fail-over mecanism.
Most of time, it works fine. If I cut (unplug) the WAN link not used by the pings flow, there's no interruption in the pings flow. But if I cut the WAN link used by the pings flow, there's a0 to 1-2s interruption in the pings flow, until the flow is balanced on the other WAN link.
But sometimes I noticed that the fail-over balance could take 40-60s.I imagine this could be in relation with the "initial-delay" parameter but I didn't understand why this delay occurs or not.
And more annoying, sometimes the fail-over balance doesn't work at all. But if I unplug and replug the operational WAN link (the one which is not cut), the fail-over balance occurs immediatly (even if th eunplug/plug delaus 1s). When the fail-over hasn't balanced on the active WAN , I checked that this link is really in an active status on both ER-8 and that the ping check sequence exists relally (by sniffer)
The last issue is the most annoying cause there's nothing worst than a backup solution which doesn't work the day you need it !
Thank you in advance.