01-18-2008 07:25 AM - edited 03-05-2019 08:33 PM
Hello People,
I have an Catalyst 2960G Switch.
I want to bundle 2 ports so it operates at 2gbit/s.
On 1 side I have an IBMServer1 with EthernetBonding also at 2gbits bundled.
And on the other side I also have an IBMServer2 with EthernetBonding at 2gbits bundled.
See the attached file for a simple drawing.
I am absolutely sure the config of the IBM servers is correct, because I've tested it with crosscables and
measured the bandwith with a linux program called 'Iperf' at 1,8Gbit/s
Her is my relevant Cisco Config
interface Port-channel1
switchport mode access
!
interface Port-channel2
switchport mode access
interface GigabitEthernet0/17
switchport mode access
channel-group 1 mode on
spanning-tree portfast
!
interface GigabitEthernet0/18
switchport mode access
channel-group 2 mode on
spanning-tree portfast
!
interface GigabitEthernet0/19
switchport mode access
channel-group 1 mode on
spanning-tree portfast
!
interface GigabitEthernet0/20
switchport mode access
channel-group 2 mode on
spanning-tree portfast
I'm able to ping from serverA to serverB so there is an connection.
It is also redundant because when I physically detach a network cable the connection stays alive.
But when I measure the bandwith from ServerA to ServerB, it says its maximun bandwith is 941Mbit/s, while is should be able to reach 1800Mbit/s.
I've also tried it witch the LACP protocol but I got te same result.
Am I missing something here?
Help is very much appreciated.
Hendrik van der Ploeg
Linux System Administrator
The Netherlands
01-18-2008 07:59 AM
Hi,
Issue an: sh etherchannel 1 detail.
Why don't you use: trunk ?
!
interface Port-channel1
switchport trunk encapsulation dot1q
switchport mode trunk
!
!
interface GigabitEthernet0/47
description UPLINK
switchport trunk encapsulation dot1q
switchport mode trunk
channel-group 1 mode desirable
!
interface GigabitEthernet0/48
description UPLINK
switchport trunk encapsulation dot1q
switchport mode trunk
channel-group 1 mode desirable
!
BR,
Bjornarsb
01-18-2008 08:00 AM
Hendrik-
The first thing I would do is set the server and the switch speeds to 1000. Auto-neg rarely seems to work on servers and Cisco switches. You can check to see if the are negotiating (in)correctly with the show interface command.
HTH
01-18-2008 09:26 AM
Etherchannel is not a load sharing function. If you have a conversation that is strictly going between 2 servers that conversation is only going to go down one of those links thus probably it is reporting 940 meg . It is all based on how you set up the load balance for the channel but a single conversation will always go down a single link of the channel .
01-18-2008 11:04 AM
Glen is right, you would need another host in similiar setup and use ipperf at the same time.
Use test ether-channel to make sure that each server will send on a different link otherwise you will only get up to 1 Gbps performance.
Also, spanning-tree portfast causes more problem than it solves with linux bonding because STP has to properly manage the MAC so it doesn't flap, but again it depends on what bonding mode you are using.
LACP should be fine, but stuff like mode=1 needs STP on for the port.
01-19-2008 04:10 AM
OK Thanks!!!
I'm back in the office on monday.
I'll give it a try then
01-19-2008 06:59 AM
Hi Hendrik
This is due to the distribution rule of cisco's
Understanding How EtherChannel Frame Distribution Works
EtherChannel distributes frames across the links in a channel by reducing part of the binary pattern formed from the addresses in the frame to a numerical value that selects one of the links in the channel.
EtherChannel frame distribution is based on a Cisco-proprietary hashing algorithm. The algorithm is deterministic; given the same addresses and session information, you always hash to the same port in the channel, preventing out-of-order packet delivery.
The address may be a source, a destination, or a combination of two IP addresses, two MAC addresses, or two TCP/UDP port numbers depending on the policy that is adopted through the ip, mac, session, and ip-vlan-session options of the set port channel all distribution command. See the "Configuring EtherChannel Load Balancing" section for detailed information
See (or other Docu's)
http://www.cisco.com/en/US/docs/switches/lan/catalyst6500/catos/8.x/configuration/guide/channel.html
Groetjes
Willem
01-21-2008 12:59 AM
ok i've given the command: show interfaces port-channel 1
At the bandwith section is says 2000000kbit.
So I guess the configuration should be fine.
But what is the best option at loadbalance?
And how can I test the both etherchannels at the same time.
The "test ether-channel" won't co-operate with me
test ether-channel load-balance interface port-channel 1 ip 10.168.1.28 255.255.255.0
Then it gives back:
Would select Gi0/17 of Po1
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide