Optional IS-IS Fragment Timestamping

Many applications in today’s networks rely on reliable and timely flooding of link-state information, such as, but not limited to Traffic Engineered networks and advanced telemetry solutions. If such information is delayed during flooding it can be difficult for those applications to adequately fulfill their intended purpose. This document describes extensions to ISIS allowing it to carry the origination time on each fragment. The origination time can be used to aid troubleshooting of large domains and/or by the applications themselves to improve their behavior. As an example, in case of Traffic Engineered Networks synchronization of the Traffic Engineering Database (TED) enables the compute nodes to adapt to changes in the network state and/or react to network events in a timely manner. Relying on a synchronized TED while the flooding information is delayed can easily lead to service degradation due to substandard re-optimization of network load. The origination time proposed in this document is meant to be used by the compute nodes or by an operator of Traffic Engineered Network to measure any delays incurred in TED synchronization. The awareness of delays in the distribution of information can be incorporated further into algorithms and network tooling to improve the responsiveness and quality of decisions taken.

This section defines a new, optional TLV that can be present in any fragment. In case of multiple instances of the TLV in a fragment only the first occurrence MUST be used. The semantics of the TLV is the point in time the fragment with the current sequence number has been generated. Its absence signifies that such information is not available due to host of possible issues, one of them lack of clock with synchronization precise enough. For practical purposes, although desirable, timestamping the moment a fragment is flooded would be preferable but beside practical implementation problems this could generate on different interfaces the same fragment with different content which breaks one of the fundamental tenants of link-state protocols. However, an implementation is free to choose to use, e.g. the moment the fragment is queued for flooding first time rather than the time the version is generated. To save space the timestamp is following semantically NTP seconds epoch with the exception of an extra bit in the seconds field to extend the wrap around and carrying only 2^-8 of a second as maximum resolution of the timestamp since this is considered sufficient for link-state purposes. The specification follows further guidelines of as far as possible.

Type: TBD1
Length: ...
Seconds: 4 bytes of number of seconds since the NTP epoch.
H(-Bit): 1 bit. Extra high order bit is used to prevent wrap-around in 2036 and pushes it out to 2242. The offset can be constructed in network order `HB` shifted to left without overflow by 32 bits and the `Seconds` field OR'ed into the according value.
Fraction: 8-bits of fraction of the second in units of 2^-8 which is equivalent to 1/256 of a second or roughly 4 msecs resolution.
Precision: 7 bits indicating the maximum possible slip (either in future or past) of the clock used to generate the timestamp (depending on the synchronization protocol) as 2^Precision where at minimum of the range signifies 2 msec or better precision and the maximum of the range amounts to 256 msec precision or less. A node that cannot achieve this minimum precision required SHOULD NOT advertise the fragment timestamp.

A requirement for the correct interpretation of the additions proposed in this document is an infrastructure capable of synchronizing time across devices involved so the timestamps at the various points of interest become comparable. This could be accomplished by utilizing NTP , Precision Time Protocol (PTP) IEEE Std. 1588 or 802.1AS designed for bridged LANs. The achieved precision is carried in the timestamp of the fragment. Though the timestamp can be very useful in deriving measurement of behavior in a deployed IS-IS network, e.g. maximum incurred flooding delays between any pair of nodes, it should not be used in any attempts to modify the behavior of protocol behavior itself such as e.g. influencing flooding rates. A single badly synchronized clock could otherwise change the behavior of parts or even the whole network in unpredictable or even detrimental way.