It really depends on the use-case and the implementation. This measurement may be excessive if running at a 3.3ms or 10ms interval, but you don’t run these intervals on anything but the best and most deterministic of links. For links with higher or unpredictable latency, the typical intervals are at least 50msec (typically between 100msec and 500msec). At those rates, the overhead is not significant.
At the same time, with more software implementations coming to the market, the overhead is smaller compared to the hardware implementations as there is no additional offload to a different engine. Ashesh On Dec 19, 2017, at 12:24 PM, Jeffrey Haas <jh...@pfrc.org> wrote: Ashesh, I'll take it as a given that there's an implied gripe about a lack of TLVs for BFD and a push for BFDv2. :-) The work in here seems reasonable, but does run up against the question I always must ask: Is this actually useful/usable at high BFD rates? I understand that a likely scenario (and not documented in the draft) is you start with a sedate enough transmit interval so as to get your measurements. But once you have them, I would expect that a poll would be initiated to utilize faster timers. Once we get to the fast timers, would the procedures for handling the timestamps be reasonable to run at that rate? -- Jeff > On Wed, Dec 06, 2017 at 01:52:47PM +0000, Ashesh Mishra wrote: > Hi BFD experts, > > We recently submitted a new individual contribution on BFD Performance > Measurement. Please review the document and provide comments. > > https://datatracker.ietf.org/doc/draft-am-bfd-performance/ > > This document proposes a mechanism to determine the smallest BFD transmit > interval that can be supported on the link. This is achieved by actively > measuring the one-way delay for each BFD session and setting the BFD session > intervals based on the measured delay. This allows the BFD session to adapt > to the fastest rate feasible on the current active path. > > The method described in this proposal is useful in networks where the network > latency is high, or varies with time. Trans-oceanic links and connectivity > over geo-synchronous satellites are typical examples of links where the > latency is high and the difference in latency on primary and backup paths can > be significant. Another use-case is connectivity using satellites in > mid-earth orbit (MEO) or low-earth orbit (LEO). In these systems the one-way > delay, while it is low (25msec to 150 msec), varies with time. This > variation, based on various factors, can be as high as 30 msec. With mobile > receivers, such as ships, the delay when using such connectivity can be > non-trivial to predict. This requires an automated method to determine the > optimal BFD interval to allow fastest possible recovery in case of failure. > > Thanks, > Ashesh Mishra > Mahesh Jethanandani
