I will try to explain my problem.
I want to increase bandwith to 2Gb/s on my network.
My network is really easy : 2 computer with 2 ports ethernet 1Gb/s configured in bonding (mode 0 round robin), so there is load-balancing and a better bandwidth.
Between the 2 computers, we have our cisco 4948 Catalyst. So i created port-channel 1 (port 3 and 4 for the 1st computer) and port-channel 2 (port 7 and 8 dor the 2nd computer) on mode 'ON' .
I read that mode 'ON' is enought to just increase bandwidth.
But it doesnt work, i tried other mode (desirable auto active passive), dont work too, i try to change the load-balancing : src-dst mac, ip, port, dont work. Always the same bandwidth : around 950 Mb/s
i did a test without the Switch : The computers are linked directly with 2 cross-cable, and the bandwidth is 1,8 Gb/s, Means bonding on Pc works.
PS : i'm french sorry.
Same type of traffic with the same source and destination will hash to the same link.
To notice the increase of bandwidth, you need to run different type of traffic, i.e: run a FTP session along with a HTTP download session and you can see both links in the switch being utilized - you must also use the 'src-dst-port' port-channel option.
"Same type of traffic with the same source and destination will hash to the same link"
Well it means when i do iperf, the packets are only use 1 link because of same source and destination?
Well i will try to describe my tests.
So im using 2 servers iperf on different ports (5001 on 5004) on the same computer (i open 2 windows terminal) and on the 2nd computer, 2 clients iperf.
Results : sometime i got 500Mb/s on each server, or i got 800Mb/s on each server.
I think sometime the 2 servers used the same link, or they used the 2 links.
Load-balancing src-dst port
Channel-group mode on
Your findings are correct. The hash is unpredictable as to which link is going to use and based on your findings, if you are sending 800Mbps on each iperf instance, then you are using 80% on each link.