+1 I don't even care if the safety breaker is a little late, e.g., a few hundreds of packets (as might be needed for efficient implementation). Joe On 1/13/2014 1:42 AM, Eggert, Lars wrote: > Hi, > > On 2014-1-13, at 10:16, Xuxiaohu <xuxiaohu@xxxxxxxxxx> wrote: >> No conflict at all. What I meant is: for those clients of MPLS which are not TCP-friendly (case 2&3 as described in Section 3.1.3 of RFC5405), they should never be transported over the unprovisioned path (e.g., the Internet). Insteads, they should only be transported over a provisioned path in a restricted networking environment. As a result, there is no need for the congestion control mechanism for them. > > I agree, but I think we need a safety mechanism when such traffic does end up on the general Internet (because operators may not read the RFC, or there may be configuration errors, etc.) > > Even when running inside a provisioned domain, you probably want some sort of safety net, like a circuit breaker that detects if your tunnel is experiencing/causing severe congestion, and shut it down. > > Lars >