[Detnet] Analysis related to packet ordering function (draft-varga-detnet-pof)

[Mohammadpour Ehsan <[e-mail]>]

Hello Balàzs, Janos, Stephan and Tobias,

Thanks for your efforts on the packet ordering function draft. On the related topic, I’d like to share with you and the WG our recent work on packet reordering in deterministic networks. Here is the link for the full paper: https://arxiv.org/abs/2008.03075. The paper provides formal analysis to obtain the parameters of re-sequencing buffer (as a POF mean), i.e., time-out value (i.e. POFMaxDelay) and buffer size, and placement impact of the re-sequencing buffers on worst-case delay and delay-jitter[1]. The main discoveries of the paper are as follows.


  *   The time-out value of a re-squencing buffer (i.e. POFMaxDelay) is equal to the bound on a per-flow metric, called "reordering late-time offset” (RTO) as in RFC4737.
  *   The size of re-sequencing buffer is equal to:
     *   the bound on another per-flow metric, called "reordering byte offset” (RBO) as in RFC4737, when the network is assumed to be lossless (by enabling packet replication methods).
     *   a function of flow arrival curve[2] and the time-out value, when there might be packet losses in the network.
  *   How to obtain the bounds on RBO and RTO? In the paper, we also compute tight bounds on these two metrics; specifically,
     *   The bound on RTO is a function of the flow-path delay-jitter and arrival curve of the flow.
     *   The bound on RBO is a function of arrival curve of the flow, flow-path delay-jitter and the time-out value of the re-sequencing buffer.

With the aforementioned findings, one can dimension the sizes and time-out values of the re-sequencing buffers, placed for DetNet flows, to provide in-order packet delivery.

Moreover, the paper studies the impact of placing re-sequencing buffers on delay, delay-jitter bounds, and flow burstiness. In short,

  *   When the network is considered lossless (e.g. by means of packet replication), placement of re-sequencing buffers does not increase the worse-case delay and delay-jitter. This for-free property of re-sequencing buffers leads to the in-order delivery of packets while preserving the same guaranteed delay and delay-jitter bounds.
  *   When the network has packet losses, re-sequencing buffers increase the worse-case delay and delay-jitter as much as their time-out values.
  *   We capture the burstiness increase at the output of re-sequencing buffer as a function of arrival curve, delay-jitter of a flow-path and time-out value of the re-sequencing buffer.

Finally, the paper also discusses the placement strategy of re-sequencing buffers throughout a network, which I refer to the original paper for more details.



Best,
Ehsan



--
Ehsan Mohammadpour
PhD Candidate at Swiss Federal Institute of Technology (EPFL)
EPFL EDIC PhD student representative
IC IINFCOM, LCA2, INF 011, Station 14, 1015 Lausanne, Switzerland
https://people.epfl.ch/ehsan.mohammadpour


P.S.
[1] delay-jitter: it is defined as the difference between the worst-case and best-case delays of all packets of a flow.
[2] arrival curve: it is a bound on the total number of bits/packets sent from source within any period of time. It is also referred to T-SPEC in some documents.