Clone wiki

meetings / 140612_webex_otf

Minutes Webex 12 June, OTF call 6TiSCH WG

Note: timestamps in PDT.

Info

Agenda

  • 6top-OTF interaction
  • Status of the OTF draft
  • TODO for having a new complete version by cutoff date

Slides

Taking notes (using Etherpad)

  • Xavi Vilajosana
  • Thomas Watteyne

Attendance (alphabetically)

  • Diego Dujovne
  • Maria Rita Palattella
  • Nicola Accettura
  • Pascal Thubert
  • Qin Wang
  • Thomas Watteyne
  • Xavi Vilajosana

Action items

  • Pascal to prepare a 5min presentation to describe/discuss/present track management in the following 6TiSCH meeting.

Minutes

  • [22.03] Meeting starts
  • [Maria Rita] Interactions OTF / 6top
    • When OTF needs a new softcell, needs to be associated with a bundle which comes with a track
    • We have multiple tracks. We may specify the trackID or not, in the latter case the cell may belong to any track. But when should OTF ask for such a trackless cell? And if so, what is the mechanism to assign to a track?
  • [Thomas] Tracks are used to separate flows. If we are using OTF, everything is soft-cells and everything is best effort. So tracks seem to be something not needed, as we do no have the notion of different flow from a QoS point of view.
  • [Pascal] Huge confusion involving deterministic tracks set up by the PCE, and the IP link between to adjacent nodes. OTF plays on that second case. The PCE has assign hard cells. The track can be used to mix deterministic and best effort traffic. Imagine a packet in a track that for some reason finds a track that is congested; the node might use a best effort cell for one hop, but in the next hop recovers the track.
  • [Thomas] I envision an OTF flexible enough so it can be used without any central PCE.
    • OTF decides when to reserve cells according to dynamic needs.
  • [Qin] Clarification: two options:
    • associate a soft cell with a single best-effort track, or
    • associate a soft cell with multiple tracks at the same time
  • [Maria Rita] vision aligned with Pascal: cells are assigned to tracks.
  • [Qin] The CREATE soft cell command take trackid as a parameter. If this id is set to 0x00, the cell is considered best-effort.
  • [Pascal] I think OTF needs to play only on best effort tracks. Not play by now on end to end flows.
  • [Thomas] single hop algorithm. When you ask 6top to create an extra soft-cell, it is 6top which selects the exact cell.
  • [Maria Rita] despite considering OTF a simple allocation mechanism, the selected cell belongs to a track.
  • [Pascal] we decided that trackid 0x00 means best effort, outside of the flow priorities, would be used.
  • [Thomas] The concept of track in OTF is something that can be elided as all cells will be scheduled in the same best effort track.
  • [Maria Rita] OTF will always check the total BW between 2 neighbors.
  • [Pascal] variable traffic exists, we may want to change the size of a track we may want to play with the BW allocated to a track but this will require end to end traffic flow management such as RSVP.
    • In traditional IP, when there is excess traffic, frames are placed in best effort tracks in order to cope with too much demand.
  • [Thomas] if we start with the common BW on the best-effort track, it will be a very good start. Enables us to implement that for the plugfest.
  • [Qin] OTF can look at queues and decide what to do. Is that what you mean?
  • [Thomas] OTF will decide to add cells to a particular neighbor according to the size of the queue.
    • What if we have more than one queue? the concept of queue is an implementation decision, at the end there is an order that enables to sort all packets in a single queue according to their priority and destination.
  • [Qin] Node A has 3 tracks, best effort tracks. All of the packets are in a queue. track 1 has more demand.
  • [Pascal] there are no tracks. Only neighbors.
  • [Thomas] let's forget about tracks in the context of OTF.
  • [Pascal] we need to differentiate layer 3 flow management from layer 2 flow management.
    • OTF is a L3 thing, L2 plays with end to end tracks, it uses a reservation protocol.
  • [Pascal] proposes to list the possibilities in the architecture draft. It might be an enumeration to have a notion of what is possible and what we address.
  • [Thomas] we need to talk about queuing behaviors and track management. In this call however is out of scope.
  • [Pascal] use 5min in the following meeting to describe/discuss/present track management.

Updated