10-23-2008 05:58 PM
I hope that you can help me. I have 3 interfaces between one T320 and ERX1440, and all interfaces are configured into OSPF protocols and these are used ECMP.
From the side of the ERX, I can see the ECMP is working fine because the traffic of output (upstream) is balance and the traffic of input (downstream) isn't balance. Then go to the side of the T320 and I see the traffic of input is balance but the traffic of output isn't balance.
I show the monitor interfaces traffic of this T320:
Interface Link Input packets (pps) Output packets (pps)
so-0/0/0 Up 65332664448 (16708) 80915598507 (21325)
so-0/0/1 Up 65067222152 (16419) 61247997827 (6521)
so-0/0/2 Up 25816974549 (17043) 30046732512 (20763)
I enable the policy of load balancing per-packet but I can see that isn't working fine.
Could anybody say me some idea of the happening are into these routers?
Thanks in advance,
Solved! Go to Solution.
10-24-2008 12:19 AM
It is possible that you're just sending traffic that results in packets being mostly put onto two of the interfaces based on the hashing algorithms used on the T series. Are the source and/or destination addresses in a particularly small bunch? Is traffic all of one type (tcp or udp port)? This normally causes queries in a lab environment where the test generator may generate all the packets sourced and destined to a single (or small range of) ip address(es). Also, if traffic arriving at the T320 to be sent over the ECMP path is all arriving on a single interface, that removes one of the differentiators used to spread the traffic.
If none of those apply, feel free to send a little more detail about the actual traffic patterns and the config of your T320 and I'll certainly try to help.
10-30-2008 03:07 AM
This environment is not a lab, this is happenning in production. These links are 3 connections to a ERX 1440 and the type of traffic is typically traffic to internet (it's mixed of the TCP and UDP traffic).
I had configured the load balancing of the next form:
nsn@T320> show configuration policy-options policy-statement balanceo-de-carga
nsn@T320> show configuration routing-options forwarding-table
But I note that traffic it's not balancing in direction from T320 to ERX, however the traffic in sense from ERX to T320 is balancing.
I'm not sure if this configuration of "load balancing" per-packet is fine or I need to specific more things as hash-key or indexed-next-hop. And I'm not sure if configuring these parameters (hash-key or indexed-next-hop), the traffic will be balancing correctly.
Thanks in advance for you help.
10-30-2008 04:12 AM
Yes, I spotted that it was the traffic going from the T to the neighbour that was imbalanced. :-)
Is the traffic labelled (MPLS) when it arrives at the T series and is transmitted to the ERX? If so, then you may want to change the hashing to take that into account and to look deeper into the packet into the L3/L4 headers to improve hashing.
set forwarding-options family mpls hash-key label-1
set forwarding-options family mpls hash-key label-2 (if your traffic is VPN labelled).
set forwarding-options family inet hash-key layer-3
set forwarding-options family inet hash-key layer-4
10-31-2008 09:07 AM
I will try with your suggestion, only I had a question: What is the different between the commands "hask-key layer 3 and 4" and "indexed-next-hop"?
I understand that the command "indexed-next-hop" allows balancing in unicast and aggregate next-hop when there are interfaces aggregates.
But I'm not sure if these commands have the same effect.
Thanks for your help,
10-31-2008 12:29 PM
I'd have to admit I'm not familiar with indexed-next-hop. I'd have to look it up but I'm currently at an airport waiting for my flight home :-) I'll take a look and try to catch up with this on Monday.