0-RTT Converter PoC over Real 5G TCPM Working Group Meeting, IETF 105, Montreal 0-RTT Converter PoC over Real 5G KT and Tessares joint work based on draft-ietf-tcpm-converters SungHoon Seo, H. Yu, J. Choi {sh.seo, hyun.yu, choi.jinho}@kt.com M. Baerts, G. Detal, O. Bonaventure {matthieu.bearts, gregory.detal, olivier.Bonaventure}@tessares.net 2019. 07. 25 Korea Telecom

The 0-RTT TCP Convert Protocol 1 The 0-RTT TCP Convert Protocol https://tools.ietf.org/html/draft-ietf-tcpm-converters More Multipath TCP enabled clients than Multipath TCP enabled servers Clients want to benefit from MPTCP at least on a fraction of the end-to-end path Initial Motivation C Not ready for a MPTCP approach, but a generic one could be acceptable IETF Discussion

2 The Basic Design https://tools.ietf.org/html/draft-ietf-tcpm-converters Commands and responses are encoded as TLVs Ensures extensibility Commands are sent inside SYN Provides 0-RTT to minimize connection establishment delays Responses are returned in SYN+ACK A plain transport mode is used between Clients and Converters (no encapsulation) 0-RTT Convert Protocol is an application-level protocol listening on a specific TCP port Allows Clients to bypass the Converter Clients can learn TCP options supported by Servers

TLV message in SYN payload 3 Simple Example How 0-RTT Convert protocol works? TLV message in SYN payload @t C SYN SYN [Connect @s:p] Data SYN+ACK [Ok] SYN+ACK TLV in the payload Data No TLV is supplied @c @s

Proof of Concept Client side 4 Proof of Concept KT-Tessares joint collaboration for implementation and engineering based on the draft Open-source client library and wireshark dissectors released by Tessares https://www.tessares.net/technology/open-source-contributions/ Compare 0-RTT convert and SOCKSv5 protocol capable android smartphones PoC Clients : Samsung Galaxy S10 5G (SM-G977N) in ENG firmware Client side 0-RTT Convert SOCKSv5 an open-source SOCKS integration redsocks 5G only, 5G+WiFi with MPTCP 5G only 5G+WiFi

5G first, but some portion over WiFi Measurement Results Comparison with conventional proxy way Commercial 5G network (NSA, 3.5GHz NR, CUPS 5GC, IPv6) + IEEE 802.11ax WiFi Compare latency for setup connection for 0-RTT Convert and SOCKSv5 protocol TTFB : Time to First Byte; From time of SYN sent, to connection acceptance KT datacenter containing both Tessares' modified HAG supporting 0-RTT Convert protocol KT’s mobile MPTCP proxy supporting SOCKS w/ authentication (rfc1928, 1929) Test Scenario and Initial Results msec 0-RTT Convert SOCKSv5 MP* (WiFi+5G) 5G only MP** (5G+WiFi) MEAN 27.8 64.7 120.5 195.9 MIN 5.4 49.2 97.9 160.9 MAX 43.6 107.7 142.9 258.8 5G first, but some portion over WiFi All over WiFi * Primary route is WiFi ** Primary route is 5G

End-to-End Signalings 0-RTT Convert vs. SOCKSv5 w/ authentication 0-RTT Convert (1RTT+1RTT) SOCKS (4RTTs+1RTT) Converter (Client) Converter (Proxy) Server SOCKS (client) MAGW (MPTCP Proxy) Server A A TCP SYN TCP SYN + Server IP:Port TCP SYN TCP SYN/ACK TCP SYN/ACK TCP ACK B TCP SYN/ACK 1 RTT TCP ACK SOCKS Method Req 1 RTT TCP ACK TCP ACK Data Request (e.g., HTTP Get) SOCKS Method Res TCP ACK Data Response (e.g., 200 OK) C SOCKS Auth. Req TCP ACK SOCKS Auth. Res TCP ACK SOCKS Connect Req TCP ACK TCP SYN TCP SYN/ACK SOCKS Accept B TCP ACK TCP ACK 1 RTT 4 RTTs Data Request (e.g., HTTP Get) TTFB = B – A // Time To First Byte TTGTFB = C – A // Time To Get the First Byte Data Response (e.g., 200 OK) C

Q & A