>> As a router must evaluate certificates and ROAs which are time >> dependent, routers' clocks MUST be correct to a tolerance of >> approximately an hour. > does there need to be a 'why' for the 'approximately an hour'? (like: > "since granularity of cert time-to-live's is 30 mins on the minimum > side" ... wordsmith or whatever as appropriate)
because rpki processing cycles are likely to be of that tolerance. but do we really want to go down this path? >> If a router has reason to believe its clock is seriouly incorrect, it > 'has reason to believe' ... seems hard to do i suggested two well-know ways and got nit picked within 42 seconds. i would be happy to put the examples, 1970 and 2005, back. but they are just lint attractors. > Would it be appropriate to hack in a timehack to the rkpi-rtr > protocol? (featurecreeper!) no. we will not reinvent time synchronization. >> Servers SHOULD provide time service, such as NTP [RFC5905], to client >> routers. > it seems that 'SHOULD' here may lead into forcing (or appearing to > force) SHOULD != MUST randy _______________________________________________ sidr mailing list [email protected] https://www.ietf.org/mailman/listinfo/sidr
