2960 Ether Channel Loadbalancing issue

[manohar.yadavs]

I see one of the ether channel member interface: Gi2/0/51 is taking more traffic and is showing incrementing packet drops. The other member interfaces are not having as high traffic and no incrementing packet drops. The load on all the interfaces is showing as "0x00"(Zero). Is there any issue with the loadbalancing on port channel?

#show etherchannel summary | beg Po6
6 Po6(SU) - Gi2/0/51(P) Gi2/0/52(P) Gi3/0/51(P)
Gi3/0/52(P

#show etherchannel 6 detail
Group state = L2
Ports: 4 Maxports = 8
Port-channels: 1 Max Port-channels = 1
Protocol: -
Minimum Links: 0

Ports in the group:
-------------------
Port: Gi2/0/51

Port state = Up Mstr In-Bndl
Channel group = 6 Mode = On Gcchange = -
Port-channel = Po6 GC = - Pseudo port-channel = Po6
Port index = 0 Load = 0x00 Protocol = -

Age of the port in the current state: 524d:07h:26m:22s

Port: Gi2/0/52
------------

Port state = Up Mstr In-Bndl
Channel group = 6 Mode = On Gcchange = -
Port-channel = Po6 GC = - Pseudo port-channel = Po6
Port index = 0 Load = 0x00 Protocol = -

Age of the port in the current state: 524d:07h:26m:22s

Port: Gi3/0/51
------------

Port state = Up Mstr In-Bndl
Channel group = 6 Mode = On Gcchange = -
Port-channel = Po6 GC = - Pseudo port-channel = Po6
Port index = 0 Load = 0x00 Protocol = -

Age of the port in the current state: 524d:07h:27m:48s

Port: Gi3/0/52
------------

Port state = Up Mstr In-Bndl
Channel group = 6 Mode = On Gcchange = -
Port-channel = Po6 GC = - Pseudo port-channel = Po6
Port index = 0 Load = 0x00 Protocol = -

Age of the port in the current state: 524d:07h:27m:48s

Port-channels in the group:
---------------------------

Port-channel: Po6
------------

Age of the Port-channel = 524d:07h:25m:01s
Logical slot/port = 9/6 Number of ports = 4
GC = 0x00000000 HotStandBy port = null
Port state = Port-channel Ag-Inuse
Protocol = -
Port security = Disabled
Load share deferral = Disabled

Ports in the Port-channel:

Index Load Port EC state No of bits
------+------+------+------------------+-----------
0 00 Gi2/0/51 On 0
0 00 Gi2/0/52 On 0
0 00 Gi3/0/51 On 0
0 00 Gi3/0/52 On 0

Time since last port bundled: 524d:07h:23m:46s Gi3/0/52

#show etherchannel load-balance
EtherChannel Load-Balancing Configuration:
src-mac

EtherChannel Load-Balancing Addresses Used Per-Protocol:
Non-IP: Source MAC address
IPv4: Source MAC address
IPv6: Source MAC address

[MHM Cisco World]

Change the hash to be scr-des-mac

Also you need to change the softmax

 qos queue-softmax-multiplier 300

[manohar.yadavs]

Should I change the Hash on both the peer end switches? or just this switch?

[MHM Cisco World]

Yes change in both SW

[Joseph W. Doherty]

"Is there any issue with the loadbalancing on port channel?"

Yup, there sure is!  You want to use the load-balancing algorithm that "best" distributes traffic across your links.

Choices that include both src and dst often work well, such as src-dst-mac suggested by @MHM Cisco World , although I generally suggest using src-dst-ip, if device supports.  Using a -mac variant only considers distribution on the same L2 segment, so, for example, Etherchannel between a pair of L3 switches, on a routed port-channel link, only "sees" those L3 switches MACs, but using -ip load balancing can also "see" the IP variety both local to same L2 segment and all L3 transit traffic.

MHM makes another good suggestion, increasing qos queue-software-multiplier to 300, but 1) I recall Cisco recommends bumping this value to 1200 (if you have drops, which you might not have after changing your load balancing algorithm choice), and 2) that command might not be supported on an older 2960 whose QoS architecture is like a 3560/3750, not like the later 3650/3850 switches which I believe use that command.

Lastly, MHM recommends changing both port-channel switches, which when using, again, a src-dst variant is probably okay, but also again, there are time when the two sides should use different load balancing choices.

[MHM Cisco World]

LoL... I know it you will mention the use of 300 instead of 1200
I know it 
OK I will provide cisco doc. about the soft max multiply value for each SW. 
have a nice day friend  @Joseph W. Doherty 

[Joseph W. Doherty]

LoL... I know it you will mention the use of 300 instead of 1200
I know it 

Ha, then you know me well, as I do read your postings (always looking for some new tidbit of information I might learn).

OK I will provide cisco doc. about the soft max multiply value for each SW. 

For a 2960/3560/3750, basically there's a similar way to do the "qos queue-software-multiplier" function but totally different syntax, i.e. "mls qos queue-set output # threshold # # # # #"