Question

Share does not loadshare as it should


Hi
Im creating a link aggregation with LACP between a X670V and a Ericsson SSR8000. The share contains 4 10Gb links and works fine except that it doesn't loadbalance over all 4 links. We have configured it to use L3-L4 as algorithm and the destinations ipaddresses contains a fair ammount of different addresses and tcp-ports. Whatever we do it just shares over 2 ports sending out from the 670. If i disable on port that actually sends traffic then it starts using all three ports but not balanced. Anyone out there with any ideas?
BR Anders S

9 replies

Userlevel 7
Can you try using custom and crc, see if it improves the load balancing?
Userlevel 4
Check hashing and see if changing to CRC helps. Two options are xor or crc Note: multicast traffic always takes same link. Broadcast as well . Unknown unicast takes same link .
Hi and thanks for your answers.
I tried all options and i get no change still just loadsharing between 2 ports. The traffic is not multicast, only unicast with known destination.
Userlevel 4
If you disable the two links where you see traffic. Does it load share with other two active links .
Userlevel 6
Hey Anders is the traffic from the server using all 4 links? When you do a show sharing are all of the ports on the switch active?

P
The incomming traffic comes on all 4 links and all ports are active. If i disable 1 of the active ports the trasffic is shared among the 3 remaining ports but not even. It looks like if i have a even amount of links the sharing is done on half of them and if i have odd number of links it shares on all links but not even. State Speed % bandwidth % bandwidth % bandwidth % bandwidth ================================================================================ 1 R 0 0.00 2.92 0.00 49.83 2 A 10000 2.03 3.01 53.13 55.13 24 R 0 0.00 3.00 0.00 37.34 25 A 10000 2.06 4.34 0.01 55.79 ================================================================================ State Speed % bandwidth % bandwidth % bandwidth % bandwidth ================================================================================ 1 A 10000 1.48 2.92 20.33 49.83 2 A 10000 1.34 3.01 13.75 55.13 24 R 0 0.00 3.00 0.00 37.34 25 A 10000 1.38 4.34 21.04 55.79 ================================================================================ State Speed % bandwidth % bandwidth % bandwidth % bandwidth ================================================================================ 1 A 10000 1.02 2.92 27.44 49.83 2 A 10000 1.03 3.01 0.01 55.13 24 A 10000 1.07 3.00 27.95 37.34 25 A 10000 1.04 4.34 0.00 55.79 ================================================================================ * sitehub79.33 # show sharing Load Sharing Monitor Config Current Agg Ld Share Ld Share Agg Link Link Up Master Master Control Algorithm Group Mbr State Transitions ============================================================================== 1 1 LACP L3_L4 1 Y A 1 L3_L4 2 Y A 1 L3_L4 24 Y A 2 L3_L4 25 Y A 2 [/code]
Userlevel 4
This is an expected behaviour lag algorithm doesn't do equal load balancing on all links but instead it will only load share.
Yes i know that we can't get perfect balance but i tink it shoud share among all links if you have a enough destination addresses/ports in the data flow. So i don't think it's expected behaviour.
Userlevel 4
Raise it to extreme TAC to get confirmation With the results . If it is a bug they will fix it

Reply