Are out-of-order frames really an issue on most LACP LAGs? Since most vendors use a hash on L2, L3, or both headers, any given flow would take the same physical link, correct? I have seen multiple service providers refuse to build LAGs on paths with unequal latency out of fear for this issue. As a result, they will avoid using diverse fiber routes (reducing redundancy.) If they have to use diverse paths, they will create multiple L3 links and use multipath or similar to balance traffic out. This seems like it could be worse for out-of-order frames since depending on the hardware, path selection might be more finicky.
So in that case, it seems using multiple L3 paths would either pose the same or greater risk of being out-of-order than if they were LACP, but never less. I imagine this would vary by vendor, but it seems you would always be safer just using a LAG as nearly everything uses some hashing on the headers.
Do you have policies about using diverse / unequal latency paths for LAGs? Has anyone ever actually seen problems with that kind of configuration?
No comments:
Post a Comment