From pmle at memphis.edu Tue Oct 15 11:47:16 2024 From: pmle at memphis.edu (Peter M Le (pmle)) Date: Tue, 15 Oct 2024 18:47:16 +0000 Subject: [Ndn-interest] Unicast Ethernet Face vs Unicast UDP Face (Wi-Fi) Message-ID: Hello, I am working on wireless multi-hop forwarding over Wi-Fi ad-hoc links using NDN over unicast Ethernet faces. However, I am running into a problem. This scenario should be able to be reproduced by using ndnputchunks and ndncatchunks from the ndn-tools suite over a Wi-Fi ad-hoc link (IBSS mode @ 5.240 GHz and 20MHz bandwidth (HT20)). When using ndnputchunks/ndncatchunks with Ethernet unicast faces, my throughput is very unstable and averages to 40 Mbps whereas the same setup over UDP unicast faces gives me an average throughput of 100 Mbps and is fairly consistent with that number. To setup the unicast Ethernet face, I use the command: 'nfdc face create remote ether://[a3:90:ee:ec:f0:22] local dev://wlp5s0 reliability on' where the ethernet address is the MAC address of the producer's Wi-Fi interface. To setup the unicast UDP face, I use the command: 'nfdc face create udp://10.0.0.3' where the IP address is the IP address of the producer node (statically assigned for testing). I turn off the content store on both consumer and producer to disable caching: 'nfdc cs config serve off admit off' I manually set the next hop for the name prefix using the new face id ('nfdc route add /test/data [face_id]'). I then run 'ndncatchunks -f /test/data > ./10MEG' to send an interest for a 10 MB file. Taking a look at both Ethernet and UDP transmissions using Wireshark with another Wi-Fi interface set to monitor mode, I note a significant difference in that the Ethernet transmissions use NDNLP whereas UDP does not use NDNLP. UDP uses IP fragmentation instead. Another difference is that UDP uses A-MSDU while Ethernet does not. Both, however, seem to use A-MPDU and BlockACK for aggregating Interest packets. This leads to the question of whether NDNLP is the cause of this disparity or just a victim of more fundamental concepts in Wi-Fi links. And if NDNLP should be the cause of this difference in performance, are there any explanations for which parts of NDNLP are responsible? Insights on the physical and link layer would be greatly appreciated as well. Sincerely, Peter Le -------------- next part -------------- An HTML attachment was scrubbed... URL: