|
When compiling a 3.3 kernel, I noticed that a new driver called teaming was added to the networking system. According to the relevant commit teaming is a userspace-driven alternative to bonding. Has anyone been testing this out? Is it faster or better than the old tried-and-true bonding driver? What would be the advantages of changing? |
|||||||||
migrated from serverfault.com Apr 11 '12 at 0:06This question came from our site for system and network administrators. |
|||||||||
|
|||||||||
|
It looks like the advantages of changing right now are "none at all" inasmuch as the project has only just been added to the kernel, has very little documentation, and is self-described as being "still in its dipers[sic] atm". In the long run, a userspace networking bonding driver could have some of the same benefits that FUSE (the userspace filesystem interface) brings to the world of filesystems -- primarily that it's much easier to develop and experiment with different policies, protocol implementations, and so forth. By simplifying the in-kernel code and pushing the complexity into userspace, you can also end up with a solution that is more robust in the event of failures and that allows for more agile responses to bugs and feature requests and so forth. This presentation (warning:PDF) describes the motivation and goals of the project. Primarily, they're looking to replace the legacy bonding code which is bloated and complicated with something that is smaller, easier to maintain, and more performant. |
|||||
|
