Clone wiki

meetings / 140319_webex_otf

Webex OTF authors, March 19, 2014

Attendance

  • Maria Rita Palattella
  • Diego Dujovne
  • Nicola Accettura
  • Thomas Watteyne

Goal

This meeting was scheduled in order to discuss specific items for the development of the OTF draft and the interaction with 6top and other components of the 6TiSCH stack. Both the Note Well and the Reminder sections are included in the call at the beginning.

Agenda

  • 6top & OTF (Maria Rita Palattella)
  • OTF Interface (Diego Dujovne)
  • OTF Open Issues (Nicola Accettura)
  • OTF triggering events (Alfredo Grieco)
  • AOB

Slides

Action items

  • Maria Rita Palattella to discuss the need for cell selection algorithm in 6op-sublayer draft after the negotiation phase.
  • Maria Rita Palattella to send general comments about 6top-sublayer to Qin/Xavi on 6TiSCH ML
  • Nicola Accettura to discuss the use of chunks with soft cells on the ML
  • Maria Rita Palattella discussing the relationship between tracks and soft cells reserved through OTF
  • Thomas Watteyne to discuss extensions of the YANG model for optional modules e.g. OTF. (proposal: coap://[aaaa::1]/6t/ext/otf/alg)
  • Nicola Accettura to discuss simplification of pre/post allocation on ML

  • 6top & OTF (Maria Rita Palattella)

    • Maria Rita presents a set of slides defining the relationship between OTF and 6top.
    • The first discussion item is whether OTF can allocate soft cells exclusively.
      • Even though bundles are mentioned in the 6top draft, for OTF purposes, the only meaning is to generate a group of more than one allocated cells.
    • The second item under discussion is the absence of a process to select soft cells after the negotiation phase between 6top to 6top of two neighbours.
      • Thomas recommends to discuss this item with Qin and Xavi.
    • The third discussion item is Cell collision: The 6top layer cannot differentiate collisions from other sources of packet loss. The use of Chunks may solve them, but Cells within Chunks are reserved as hard cells, thus OTF should not use them.
      • Nicola proposes that Chunks could be used either for DeTAS and for OTF.
      • Thomas comments that Chunks are still unclear but they should work in centralized mode (PCE). This matter is still under discussion.
      • Diego asks if there is call to reserve Bandwidth instead of soft cells on 6top. It also provides an example about the use of cell statistics: If a BW of 150Kbps is required and each soft cells supports 100Kbps, 6top can use 100Kbps of Cell1 and 50Kbps of Cell2, both with 100% PDR, or, if both Cell1 and Cell2 have a 75% PDR, then 6top must reserve Cell1 and Cell2 with 75Kbps, totalling 150Kbps. Does 6top take this decision, or should it be OTF? Response:?
    • Maria Rita believes that this should be part of the soft cell selection procedure that right now it is missing in 6top.
    • Maria Rita presents the lack of procedure in 6top to select the cells to be added or deleted from the schedule. Where should it be defined?
    • Maria Rita shows the use of 6top monitoring commands and 6top statistic commands, to provide input to OTF to estimate more or less BW requirement. She also describes that the policy to generate Create/Delete soft cell commands is out of 6top scope but OTF provides one.
    • Diego comments about the requirement of cells to belong to a track
    • Thomas proposes that, since tracks were created to be compatible with a centralized approach (using a PCE), a single track ID can be used for all the cells with OTF, as a starting point.
  • OTF Interface (Diego Dujovne)
    • Diego presents OTF interface requirements and issues. He proposes to use the YANG data model for algorithm selection and parameter passing, while leaving a direct connection with no intermediate layer to 6top.
    • Diego also describes the data structure and the functions needed to identify the algorithm and pass the parameters.
    • After a question from Thomas, asking if this supposes a centralized algorithm selection and parameter passing process, Diego comments that, a centralized distribution approach could be available as a start, OTF<->OTF communication will be needed when scheduling is completely distributed.
    • As a consequence, Thomas proposes to use IEs to transmit the algorithm selection and parameters between nodes when a new node arrives: a GET command can be issued to receive an (extended) IE with the required information through 6top.
    • Nicola comments that another way to distribute the algorithm is to do something similar than RPL with Objective Functions.
    • Thomas proposes also to standardize algorithm identifiers by IANA.
    • Finally, following a short discussion with Alfredo at the IETF meeting, Diego proposes to add a recommendation on the OTF draft to implement lightweight and stateless scheduling algorithms due to constrained resources available. Furthermore, he proposes to use only data provided by 6top instead of storing data from the current run to the next one.
  • OTF Open Issues (Nicola Accettura)
    • Nicola describes several OTF open issues. The first item is similar to the one of the points previously discussed by Maria Rita: OTF only allocates soft cells, since bundles are automatically created after more than one cell is allocated between two neighbours.
    • He proposes to use a single allocation policy comprising the hybrid one; Maria Rita comments that the existence of three policies is required providing options to attach to any of them.
    • Diego asks if the pre-allocation policy should be strict: preallocate only at the beginning and any other allocation (even over-provisioning) after this turn this policy into hybrid.
    • Maria Rita recaps the meaning of pre-allocation policy, showing to Nicola that it does not imply a static behavior.
    • Nicola proposes to obtain input parameters from the 6top layer, for example, using trickle. This is an alternative to the previous discussion (from Thomas) of using IEs and a centralized distribution as a start (from Diego).
    • Finally, Nicola discusses the use of Chunks in OTF: allowing them in distributed scheduling and enabling the use of soft cells; this could be discussed in the ML.
  • OTF triggering events (Alfredo Grieco)

    Alfredo Grieco could not attend this call. This discussion is pushed to a future call.

  • The call finishes by a concluding remark from Thomas, asking about the possibility to implement OTF either by simulation / hardware implementation / Ptolemy model.

Updated