Just a strange problem I’ve found on the network. Last weeks, after replacing a RB450G for a RB1200 as the core router for the network, users are complaining that network speed was much lower than usual. Doing some testing, we saw ping times to the router (within local subnet) jumping everywhere between 1 ms and 20 ms. The setup consist of a gigabit switch connected to port ether9 of the RB1200. Traffic passing the interface is between 8 and 12 mbit continously in both directions.
These are the things we’ve seen…
When connecting a test laptop (192.168.3.x) and pinging the router (192.168.3.1) on local subnet through the switch we get everything between 1 and 20 ms RTT.
After disconnecting all other devices on the subnet, no traffic passing to the router, ping times to the router are a rock solid <1 ms.
When connecting the devices back to the switch and traffic is again flowing through the router, ping times are back between 1 and 20 ms.
Ether9 is in a bridge called LAN. This morning, I’ve added Ether8 to this bridge and swapped the connection from Ether9 to it. From that moment, ping times, even with all normal traffic flowing the router, are good, all <1 ms or 1ms. Next test was for Ether10 (WAN connection), swapping it to Ether7 and the problem went away too.
Are ports Ether9 and 10 different from Ether 6, 7 or 8? I know ports 1-5 are in a separate switch to the CPU, but how about the rest? Does anybody has seen this kind of behaviour?
There is known issue with ports 9&10 on RB1200. It apparently only exhibits issues when running BGP and has apparently been fixed - if I need more than 8 ports, I now use an rb1100ah
I’ve experienced similar problems with ether9 & 10, they’re connected to the CPU differently (via PCI-e apparently) & this seems to hamper performance. Therefore I’d avoid using ether9 & 10 if possible.
Am experiencing the same with port 9 on an RB1200. The port is flapping and getting a 87%
at times 90% packet loss. Am seeing port flaps twice every minute. Is there a remedy to this?
The reason we got the 1200 is to get extra ports but this is a let down to us.
I don’t have a RB1200 but I just found this problem with a PPC460GT board I am working on.
I noticed this problem on this blog while researching this PCIE problem with the PPC460GT.
The problem with the two PCIE to gigabit ethernet controllers 9 and 10
may be the clock levels being feed into the PPC460GT chip. The normal
clock levels for PCIE are called HCLS ( 400 mv - 700mv).
The gigabit ethernet controller are probably HCLS clock compatible.
The PCIE clock levels for the PPC460GT PCIE CLOCK inputs are LVDS ( 1000mv - 1325 mv).
On the RB1200 the quad diffferential PCI Express Clock source chip U301 has 4 sets of resistors connected.
Coming directly off the clock outputs there are four 33 ohm resistors connected to either 50 ohm or 150 ohm
resistors.
If you have four 33 ohm and four 50 ohm resistors that is a problem.
You need to determine which set of 50 ohm resistors are going to the PPC460GT PCIE CLKS and change those resistors to 150 ohms. That will make those two PCIE clock LVDS levels which the PPC460GT requires.