This question is in regards to Dell 11th generation hardware that have 4 embedded interfaces (e.g. R610, R710, R810) running RHEL 5 and RHEL 6 servers. We normally run these in an active/passive mode (sometimes we use link aggregation when needed) - what I'd like to know is, is there a formal document and/or whitepaper on channel bonding best practices? For example, is it better to bond the first and second interface, or is there a performance improvement by bonding on the first and third interface. There might be a spec sheet on this - not sure - hoping that someone else has this information.
Thanks in advanced.
While there are two ports per bus on the motherboard, separating them is not going to have any noticeable performance difference, there is more than enough bandwidth for both ports to fully operate on the same bus and issues with performance are more likely to be caused elsewhere in the network chain. The hashing method being used will also have an effect on the performance depending on how it splits traffic, typically it is by TCP/UDP port and not on a per packet basis and so the split is not 50/50 between the ports. There is not a specific method that is best for everyone, so finding the best solution for your environment is the best way. Here is the user manual for the Broadcom NICs however with Linux the bonding can be set up differently. ftp://ftp.dell.com/Manuals/all-products/esuprt_ser_stor_net/esuprt_poweredge/broadcom-netxtreme-adap...
Thanks for the info. I'll take this info and read up on it a bit. Just wanted some sort of explanation and this appears to have been sufficient enough to validate what I need to do in my environment. I'll take a look at the manual and read up.