IP: referring to IP of the storage.
DK: how far away from data center, how much latency would that introduce compared to storage at the edge?
IP: Depends on data. For some data, enclaves (putting data in trusted exec. environment).
LZ: Mitigate challe nges. You cannot upload data being produced. Proposal is to address data moving challenges, whoever produced data would have their own data moving policy.
CT: ...economics question
DO: Unintentionally ignored: data provenace. Hard enough in centralized systems/databases. Edge stuff making this problem much harder. Potential critical thing.
KK: If you have a general ICN protocol stack/system that can handle confidentiality/access contro, addtiional problem incentives and bussiness models, who going to control which parts of the system. How to name stuff, whether to change it between providers, etc. difficult problems.
ES: Data generated functino of app. Metadata is data too. Lineage & provenance of data. ICN can be helpful to make metadata the data; knowledge plane.... We would want to do analytics on metadata.
DK: why cant we just do it at application layer? opposite of edge computing. There are good arguments why ICN approach; granular access to data, security, optimizations... Need to discuss
Ack Berman(?): Are you assuming greenfield deployment vs. running over-/underlay.
IP: Still at design stage
LZ: THis is still an idea. This can be either greenfield or adopted by operators
CT: Should we have "storage neutrality"?
DO: While we would like to treat storage as everything else, may be challengingg. "garbage in, garbage out" vs. "garbage in, garbage stays" (network vs. storage). "Fairness" already hard for caches
Yves Daoust Martin Benoit
(patent merging ICN & blockchain)
LM: Incentives for using blockchain? In bitcoin - generation of revenue, what's here?
MB: To provide proof of origin, store all signatures. As you bring producer node within ISP, ISP can start monetizing infrastructure by allowing producer to upload to infrastructure. ISP will have proof of delivery.
YD: Potential use case for edge nodes with storage incentivized to participate.
KK: is it proof of work?
MB: proof of origin and proof of delivery. ethereum smart contract.
IP: is it open permission-less blockchain used by content publisher/ISP. or not?
App namespace, network namespace
Transport service: user-space app for producer servcies; consumer servcies: segmentation, auth, integrity, naming, fetching, reassembly, verification, congestion control
Producer socket, consumer socket
Transport manifest (indexing, signature verification, intefrity verification, performance)
KK: is that manifest visible
LM: internally. app express intent whether to use or not the manifest
Legacy BSD socket API vs. modern post-socket API
Performance evaluation for single-core
DO: "congestion" is source-limited (not link limited). Scenarios don't demonstrate anything about congestion control
DO: do you limit size of manifest to MTU? is it merkle tree?
LM: yes, ~50 packet per manifest. No merkle (root manifests). Cost coming from SHA computation.
DO: difference per-packet performance because of signatures, not sha.
LM: also producer side differences. additional latencies while data production. Batching effect.
AB: For mapping of names, for video
LM: linear: 128 bits v6 prefix per chunk. recycle the same buffer
AB: from app perspective. mapping addresses to names.
LM: 50 mpeg-dash channels. buffer per channel; bind socket.
LZ: mobile part... how do they find address to attach to. HOw different clients using
LM: active speaker channel. managed by application layer. every time, schedule to fetch an active channel.
MM: for things like interest retransmission, congestion control. Where these functions leave in stack pictures.
LM: inside transport library
open source code in next few months
RR: Ravi Ravindran
DO: what's session
AyAz: video conferencing session. communicating with producer and moving. Application session, not related to transport layer
DO: how do you congestion control these things? general approach in ICN, CC done on interest messages. How do you CC for push?
AyAz: not impementing right now. Limited to edges of the network.
LZ: how do you know you're at the edge?
RR: going inside
DO: how do assign links to arriving data vs. closed loop congestion control. Giving priority to control messages. These are not control messages.
RR: amount of data coming of this is expected much less vs. data crossing.
DO: how do you decide whether theya re not injected by bad ppl
RR: not solving security problem here
MJM: this is a single mobile operator. Q is what happens if one outside the network? WOuld it work if outside. Don't think it works across.
RR: two layers of controller... Works.
LZ: you going to cross provider borders. cannot be pure local
RR: when u crossing admin boundaires, you need to reauthenticate.
LZ: put consumer ID everywhere
RR: scope limited to local. don't carry everywhere. Only data coming back...
LZ: basic change to ICN. what is this ID
RR: device ID to track a device.
DO: Operating in disaster scenarios. THis approach doesn't thought if local communcation possible in such scenarios. Is there are fallback mode?
RR: not breaking normal interest data.
DO: producer mobility doesn't seem to work at all without backend being up
RR/AyAz: if ther are traces in the network... info can be caches within the network...
LM: not supposed to support ancnorless?
RR: still "anchorless"
anchor is thing that knows the binding. state in the network that knows certain things..
LM: there is no anchor in data plane but thre is in control. Mapping system.
Similar to LISP where u need a mapping system in data plane. Properties similar to LISP
Aytac Azgin, Ravi Ravindran
Q: (Oran) aggregation limited to edge - do yo have data to support this?
Q: if an attacker inserts data with an handcrafted Bloom filter - what happens?
Summary of AR/VR activities in ICN
Shown previously at CEBIT 2018
Q; (Luca) Are they all showing the same thing
A: no each camera has a different view so viewers have a unique views
Q: (Eve) What is the implementation?
Smart city context, multi sensor environments
Network as “ database” - what are the APIs - fixed names may not be the right way; more database-like query
Q: (Oran) How is an embedded name
A: Just a name in the interest packet
Q: (Dirk) Fuzzy name attractive to send scoped requests but it adds complexity to the forwarding of the information itself - have you thought of ways to do that (application layer, other?)
A: I don’t know; this is why I am here. We do not want to impact the interest packets but send information that is connected to it
Q: (Oran) What is fuzz? Why don’t you consider bounding boxes?
A: We could do that too. We just want to get information from that approximate location.
Q: (Mosko) Project with bounded boxes - namespace that you can query in many way - army has a lot of work on anomaly and the data could be sent if needed?
A: Nacho’s work on CCNx naming - how to use directories - goal of the current work is to go beyond army but also IOT for social good
Q: (Borne) Query for attributes etc.?
A: Metadata and how you get it - not all in the names - expose some of the information - steer/toute based on attributes
Q: (Mosko) Tagnet/on-order information could be useful?
A: We may need to come back to that
Q: (Lixia) If you want the box Carspeak Dina Katabi *http://groups.csail.mit.edu/netmit/wordpress/projects/mobile-networks/carspeak/ - attribute based routing - at what layer?
A: Will look into it Carspeak - also layer question is in the backup slides
Comment (Mosko) SRL did work on attribute based routing
Video demo well received.
Comment: (Luca) nice to share the video priortitization/settings on the list.
Anil Jangham et al.
Description of the draft.
Bring QOS is ICN beyond interest into data.
Q: (Oran) Multiple consumers asking for different data? What do you differentiate? Quadratic problem (product of user + data requirements)
A: Data and request are used
Q: (Luca/Lixia)What is the QID? In the name?
A: Not in the name; DSCP code in the packet with additional TLV
Q: (Oran) Test question - 5 consumers with different requirements (1 now others non-important) - producer think his stuff is the best
A: Future research
Comment: (Oran) error to copy IP
A (Oran): overall to replicate IP is to kill cool congestion control for ICN (adding decoupling in ICN)
Q: (Oran) inside or inside the security envelope
A: Good question: inside the security envelope
Q: (Borje) QoS for interest msgs only because of complexity
A: (Oran) stageful forwarding you can reserve the return from the arrival of the interest message
long discussion - Dave to add comments?
Comment: (Oran )few drafts on QOS have expired - can the authors revive those now that it seems ICN will work on QoS
Q: (Mosko) how to link producer and consumer markings
Q: (Oran) the producer is the one who know which packets should be grouped together in a class; consumer knows the value of the data; get the decoupling from what QOS means for the producer from what in means for the consumer
other long discussion