Dear all: The discussion seems to lean on the assumption that the remote peer allocates the cell, which is true for child to parent.
I’ll note that for a parent to child, the frame would already indicate the next cell that the parent allocates, and the child would acknowledge that it will listen there. Cheers, Pascal From: 6tisch [mailto:[email protected]] On Behalf Of Nicola Accettura Sent: jeudi 2 juillet 2015 01:55 To: Qin Wang Cc: Michel Veillette; [email protected]; [email protected] Subject: Re: [6tisch] draft-dujovne-6tisch-on-the-fly-05 Qin, Michel, I also prefer option (2). This feature can be considered as a faster negotiation addressing "emergencies". Usually a negotiation will require two cells (1 for reservation request, 1 for reservation response). Let's call this as DEFAULT_NEGOTIATION. In the "emergency" case, we have a single cell used. I guess a single cell would be reserved, there is not enough room in the ack. Let's call this as EMERGENCY_NEGOTIATION. Actually, I think that an in-band reservation should be possible also for the DEFAULT_NEGOTIATION: a data packet is encapsulated in a 6top packet, with 6top possibly carrying the IEs related to a reservation request. IEs will be processed at the 6top layer, the payload will be dispatched to upper layers on the receiver side. Then, node B will send a reservation response to node A within another cell (in case, with the same piggybacking approach, if there is a cell from B to A and a data packet to be sent in that direction). This possibility for the DEFAULT_NEGOTIATION could save energy, since a single packet will be able to address 6top needs and deliver data packets. I point out again that this could be an option working when there is enough room in the packet to accomodate both data and 6top IEs. When there is no room for 6top IEs in packets carrying data (data occupies most of the frame space), the 6top IEs will be sent in a following dedicated packet, as in the current 6top drafts. For the EMERGENCY_NEGOTIATION, piggybacking 6top IEs with data packet is the sole option available. As you point out, the Frame Pending bit does not mean that there is an "emergency", so if node A sends a data packet to node B with the Frame Pending set and with a 6top reservation request, there is no way to discriminate if this would be a DEFAULT_NEGOTIATION or an EMERGENCY_NEGOTIATION. My question is: would that be possible to have a 6top flag bit indicating EMERGENCY_NEGOTIATION? I would also say that the reservation request IEs for softcells could be significantly reduced in size. Node A could send a subset of available timeslots (note, I'm not talking about cells), maybe in the form: starttimeslot3-endtimeslot7, starttimeslot15-endtimeslot25, meaning that timeslots 3,4,5,6,7,15,16,17,18,19,20,21,22,23,24,25 are available in its own schedule (only 2*4b ytes needed). Node A will also indicate the number of cells required, e.g. 3. Node B will answer with 3 cells: timeslot4 on channeloffset3, timeslot5 on channeloffset 10, timeslot20 on channeloffset5. In many cases the node starting a negotiation (node A) won't have to specify a preferred channel offset for each proposed timeslot (maybe will specify only a single channel offset for all timeslots proposed). This possibility would save a lot of space in the reservation request. You wrote also: "The remained issue for both of the schemes is if the time for nodeB to insert the softcell request response IE into ACK frame is too tight or not. I don't have clear answer to it." This is a very good point we need to verify. What do you think? Nicola 2015-07-01 13:17 GMT-07:00 Qin Wang <[email protected]<mailto:[email protected]>>: Michel and Nicola, Firstly, I agree It could be a good idea to let DATA/ACK frames carry softcell negotiation information, and agree it belongs to 6top-to-6top negotiation procedure, instead of OTF. Secondly, I would like to compare the two following schemes to implement the idea. (1) nodeA sends a data frame to nodeB with Frame Pending setting; nodeB reply a ACK frame, with a softcell request response IE to give nodeA additional softcell to send its following data frame. (2) nodeA send a data frame to nodeB, in which a softcell request IE is carried; nodeB reply a ACK frame, with a softcell request response IE to give nodeA additional softcell to send its following data frame. I prefer scheme (2) although it costs more bits of softcell request IE from nodeA to nodeB, because Frame Pending bit belongs to MAC layer, it indicates there is more data following, not necessarily more bandwidth will be needed. The remained issue for both of the schemes is if the time for nodeB to insert the softcell request response IE into ACK frame is too tight or not. I don't have clear answer to it. What do you think? Thanks Qin On Thursday, July 2, 2015 2:05 AM, Nicola Accettura <[email protected]<mailto:[email protected]>> wrote: Qin, you are pointing out that the bandwidth must be already provided if one wants to use the Frame Pending setting. This is what we have discussed in the past and is the state of the art in 6tisch. Michele is proposing to use the Frame Pending jointly with a temporary cell reservation in the ACK. I guess this is not in the scope of OTF, but, in case, in that of 6top-to-6top. Why? Currently 6top-to-6top could be used for allocating cells through a negotiation: 1. mote A sends a reservation request to mote B in a cell, and that signaling packet is unicast, so acknowledged in the same cell by B 2. mote B answers A with a reservation response in a following cell, which is also a unicast packet acnowledged in that same cell Michel is saying: 1. mote A sends a Frame Pending to B in a cell, and B sends the ack to A in the same cell answering with a temporary allocation So, Michel's proposal uses 1 cell to perform the negotiation that 6top does (in the current version) by using 2 cells. I guess this is the reason for defining this approach reactive. At the same time, the current 6top negotiation allows to exchange more information for reserving a common available cell. In Michel's approach I figure out many cases of failures: what happens if the temporary cell given by B to A was already busy in A (it was already part of the A's schedule, but with another neighbor, say C)? I guess A will not be able to free its memory, so leading to congestion. Right? Nicola 2015-07-01 10:09 GMT-07:00 Michel Veillette <[email protected]<mailto:[email protected]>>: Hi Qin Yes, the Frame Pending flag is defined as follow: IEEE 802.15.4-2006 section 7.2.1.1.3 “Frame Pending subfield is 1 bit in length and shall be set to one if the device sending the frame has more data for the recipient. This subfield shall be set to zero otherwise” This feature can be especially useful for the upstream traffic in a RPL DODAG. In a scenario where a DAG parent have dozens of children, dedicated timeslot will be infrequent and share timeslots suffer from contention. If a subset of these children have ongoing traffic, the parent can use the Frame Pending flag information to schedule temporary soft cells and avoid contention or speedup transfer. Michel Veillette System Architecture Director Trilliant Inc. Tel: 450-375-0556 ext. 237 [email protected]<mailto:[email protected]> www.trilliantinc.com<http://www.trilliantinc.com/> From: Qin Wang [mailto:[email protected]<mailto:[email protected]>] Sent: 1 juillet 2015 11:28 To: Nicola Accettura; Michel Veillette Cc: [email protected]<mailto:[email protected]>; [email protected]<mailto:[email protected]> Subject: Re: [6tisch] draft-dujovne-6tisch-on-the-fly-05 Hi Michel and Nicola, I think Michel's idea is interesting. But, according to my understanding, the Frame Pending setting just means there is frame following, does not mean that the current bandwidth provided by TSCH schedule, including hard cells and soft cells, is not enough to convey those frames, and then needs more bandwidth (e.g. additional soft cells) . Right? Do I miss something? Thanks Qin On Wednesday, July 1, 2015 4:03 AM, Nicola Accettura <[email protected]<mailto:[email protected]>> wrote: Hi Michel, your proposal is very interesting. However, OTF does not allocate cells directly: it just computes the estimated number of cells to add or delete into the schedule, and then sends this information to 6top. 6top is then in charge of negotiating cells among neighbors, and meybe perform the scheme you are proposing. So, your proposal seems fitting more the 6top-to-6top communication. Am I missing something? What others think about? Sincerely Nicola 2015-06-30 8:13 GMT-07:00 Michel Veillette <[email protected]<mailto:[email protected]>>: Hi Diego It’s my first reading of the “6TiSCH On-the-Fly Scheduling” (and I’m not completely done yet) and I wandering if the concept of on the fly, in a single exchange, temporary allocation of a soft cell have already been discussed. For example, a node can use the Frame Pending subfield (IEEE 802.15.4-2006 section 7.2.1.1.3) to indicate the presence of packets ready to be transmitted. Based on that knowledge, the target may add an IE in an enhanced acknowledgment to allocate a temporary soft cell (e.g. single cell). Each subsequent transmission may further re-allocate a temporary soft cell. It’s important to note that the default delay for a TSCH Acknowledgment is 1ms (macTsTxAckDelay) which seem sufficient for the processing of this new IE. This scheme is very reactive and may help dealing with non-predictable communication patterns. What do you things? Michel Veillette System Architecture Director Trilliant Inc. Tel: 450-375-0556 ext. 237 [email protected]<mailto:[email protected]> www.trilliantinc.com<http://www.trilliantinc.com/> _______________________________________________ 6tisch mailing list [email protected]<mailto:[email protected]> https://www.ietf.org/mailman/listinfo/6tisch _______________________________________________ 6tisch mailing list [email protected]<mailto:[email protected]> https://www.ietf.org/mailman/listinfo/6tisch
_______________________________________________ 6tisch mailing list [email protected] https://www.ietf.org/mailman/listinfo/6tisch
