[f-nsp] Better traffic distribution on LAG

[frnkblk at iname.com]

Thanks for the suggestion.  In this case it doesn't help because the
transport in between is a singe 10G transport doesn't preserve the LAG
that's at the Brocade's egress/transport ingress.

Frank

-----Original Message-----
From: Wilbur Smith [mailto:wsmith at brocade.com] 
Sent: Tuesday, October 20, 2015 3:55 PM
To: Frank Bulk <frnkblk at iname.com>; foundry-nsp at puck.nether.net
Subject: RE: [f-nsp] Better traffic distribution on LAG

Not sure if it would help in this situation, but you can try the command
"load-balance force-l4-hashing". It won't give you more IPs or MACs to hash
against, but it will force the MLX to take both the TCP and UDP source &
destination ports into account when calculating the hash for a L2
connection. 

This is configured in under each individual LAG:
Brocade(config)# lag blue static
Brocade(config-lag-blue)# trunk-type hash-based



Alternately, you can try per-packet load balancing; this is an older method
and won't work on routers with more than 16 LAGs though. This method just
round-robins the packets evenly across all ports in the lag, but it won't
take packet size into account. If you have traffic with lots of different
packet sizes, this may not evenly split the traffic. 

The configs is the same though:

Brocade(config)# lag blue static
Brocade(config-lag-blue)# trunk-type packet-based



The MLX also supports more fine grained hashing options, like symmetric
load-balancing across trunk ports or hash-diversify, but I haven't worked
with them heavily and dont want to recoemnd something that could
exascerbaste the problems. 

Wilbur

-----Original Message-----
From: foundry-nsp [mailto:foundry-nsp-bounces at puck.nether.net] On Behalf Of
Frank Bulk
Sent: Friday, September 25, 2015 5:46 PM
To: foundry-nsp at puck.nether.net
Subject: [f-nsp] Better traffic distribution on LAG

I'm running into an issue where our transport gear can't have their LAGs
hash on IP for egress (from Transport #2 to Brocade MLXe #2), so I need a
way generate more IPs (and MAC addresses) so I can use ECMP on Brocade MLXe
#1 to send traffic to both IPs (and therefore both MACs), and then traffic
should be load-balanced across those unique MACs. 

Here's the existing network design:

Internet
    |
Brocade MLXe #1
    |
   10G
  tagged
 VLAN 500
    |
 UNI port
Transport #1
   NNI 
  /   \
 /     \
 |      |
 |      |
 \     /
  \   /
   NNI
Transport #2
  |  |  |  |
 1G 1G 1G 1G
     LAG
    tagged
   VLAN 500
  |  |  |  |
 Brocade MLXe #2
     |
  Customer

What about VRRP on the Brocade MLXe #2, and then have a route on the Brocade
MLXe #1 to both the Master IP and Local IP that reside on Brocade MLXe #2?
There wouldn't be a Backup for the VRRP, but that doesn't matter, just need
unique MAC addresses for each Master and Local IP.  That would at least use
two of the four GigE's.

Regards,

Frank

_______________________________________________
foundry-nsp mailing list
foundry-nsp at puck.nether.net
http://puck.nether.net/mailman/listinfo/foundry-nsp