Present: Phil, Om, Rodrigo
Agenda: * To-do list for TTX * Dissemination (JHU) * Zigbee
Om: TTX: prepare some slides, materials from the paper, specifications. Phil: reach out to the community. Chating w/ David, there aren't many solid platforms, and people are implementing several protocols, like DYMO, AODV, CTP, there is no reason to not have all of these available in the tree...
Rodrigo: requiring net2 to support and write all protocols is not scalable. Maybe the role of net2 should be to provide guidelines and make writing protocols easier.
P: net2 is responsible for the code that goes into lib/net, which means maintain the code. For something to be part of the official distribution, net2 should maintain it, and that requires that they join net2.
Om: net2 could though talk about fundamental things to do when writing protocols, like how to start the radio, etc P: It would be great for net2 to create a library of components to make writing protocols easier but making people have to use them is not good. E.g., Lots of people use Click, but not everyone uses Click.
O: how about the rule that two people in the working group should be willing to be responsible for incoming code. P: The policy in Core is that responsibility means a promise that code will be supported by someone for at least 1 year P: If something falls out of support, it can be delisted. The assumption is that delisted is not being used. O: what do you think? P: If no one currently in the group is willing, and there's someone willing, the barrier would be high. R: I like the 1 year requirement for support.
O: Ok, so for the TTX we shall have a couple of slides on reaching out, and another set for past results, show the CTP run, for example. P: say MHLQI is there too
O: code freeze was? P: last week. But I don't think there are many changes.
O: Next topic: dissemination. JHU wants to join net2. Q: why does Deluge require different dissemination protocols? P: Deluge uses two kinds of dissemination: Trickle for metadata, and data. O: Can Deluge use the net2 dissemination for the metadata part? What is the difference? If their code is better, can we use it instead of the net2 dissemination? We should try to unify them P: We should ask them.
P: Kaisen Lin made some good improvements to dissemination with many items. (N.T: this is a summarized version of what Phil described)...
Trickle is fine when you have concise metadata, i.e., version numbers. Tradeoff between rate and overhead for many data items: separate trickles or one trickle where you cycle through the items. His work gets rid of this tradeoff by efficiently encoding the items that changed. Send a hash of the version numbers, and you form a hash tree to find what's different. Depending on the redundancy of the communication and the number of items that are different. Sometimes it's better to choose random items instead of going down the tree. It dynamically senses the environment, determines the redundancy and the number of items, and selects the best approach. We compared against Drip: it sent 100x more traffic, and it was only 7x faster. 14x speedup if you equalize the rates.
P: if you have more than 3 items you want to disseminate.
O: program dissemination?
P: this assumes dissemination where data items can fit in a single packet.
R: but in that case they change version numbers atomically...
P: There's been work on disseminating small, large, but not much on medium items. Maté and Tenet do it.
P: Kaisen is very interested in making this real and pushing it.
O: Next topic was Zigbee, but Andre didn't call in.
P: Are we meeting next week?
O: Should not, as most people will be away on IPSN/SPOTS.
O: Maybe we can kick off in May with the new topics, like Deluge, Zigbee...
P: CTP on the plate, tweaks for improving.