Hi John, Thanks for your comments. Please find some response inline.
Best Xuesong From: Kaippallimalil John [mailto:[email protected]] Sent: Saturday, January 30, 2021 12:56 AM To: Gengxuesong (Geng Xuesong) <[email protected]>; Uma Chunduri <[email protected]>; Lizhenbin <[email protected]> Cc: [email protected]; dmm <[email protected]> Subject: RE: [DMM] [Apn] Regarding APN Usecase in Mobile Core Hi Xuesong, The point on using GTP-u extension header for classifying each packet was that unlike classifying using DSCP (which is in the IP header), finding the QCI in GTP extension requires more searches. And the control fields in GTP-u are for signaling between the 2 GTP end points (gNB, UPF, etc). [Xuesong] According to the email from Sridhar, does "requires more searches" mean that: gNB/UPF needs to first read QFI in GTP-u extension header, then map it to the 5QI via the QFI, and determines QoS processing based on 5QI? Now with regard to DSCP, since it is a mutable field, the value can be changed on path. It may be suitable in some deployments but not as a general solution. [Xuesong] Sorry I'm a little confused here. What does "control fields" refer to in GTP-u here? The signaling is in control plane or data plane? In a gNB or UPF, there is already the subscriber context for a session and identifiers like radio bearers or IP address (upstream/downstream) to classify incoming packets. [Xuesong] "DSCP is a mutable field" may be an important limit as the interworking identifier between transport network and mobile network. But I'm not sure whether DSCP is also used in gNB or UPF, rather than transport network, for session context index? E.g., a UPF may look up session context indexed by a UE IP address, retrieve QoS, reliability, etc , classify and select MPLS label L (or a DSCP value in some cases). [Xuesong] The concept of "session context" is very interesting here. It seems like "session context" could be indexed by UE IP address and it could indicates QoS , reliability. Can I find more detailed specification about this in other documents? However, when there is a gNB or UPF function deployed in one network and IP transport is provided by another network, DSCP may not be able to convey slice aspects for a path across two different networks (this apart from DSCP being mutable). [Xuesong] Network slicing brings more complexity here. Do you think the issue of 5QI->DSCP could also be discussed together with network slicing mapping? Or we could keep them independently first? Best Regards, John From: Gengxuesong (Geng Xuesong) <[email protected]<mailto:[email protected]>> Sent: Friday, January 29, 2021 4:01 AM To: Kaippallimalil John <[email protected]<mailto:[email protected]>>; Uma Chunduri <[email protected]<mailto:[email protected]>>; Lizhenbin <[email protected]<mailto:[email protected]>> Cc: [email protected]<mailto:[email protected]>; dmm <[email protected]<mailto:[email protected]>> Subject: RE: [DMM] [Apn] Regarding APN Usecase in Mobile Core Hi John, You mentioned that "5QI/QCI etc are in the GTP extension header which may not be ideal to lookup to classify each packet in the transport network". This is slightly different from my understanding about this: 5QI/QCI belongs to control plane information, and when the base station encapsulates the packet with GTPu, it maps the 5QI/QCI into the DSCP in IP header outside GTPu encapsulation. Do I have any misunderstanding about this point? Best Xuesong From: Kaippallimalil John [mailto:[email protected]] Sent: Friday, January 22, 2021 5:25 AM To: Gengxuesong (Geng Xuesong) <[email protected]<mailto:[email protected]>>; Uma Chunduri <[email protected]<mailto:[email protected]>>; Lizhenbin <[email protected]<mailto:[email protected]>> Cc: [email protected]<mailto:[email protected]>; dmm <[email protected]<mailto:[email protected]>> Subject: RE: [DMM] [Apn] Regarding APN Usecase in Mobile Core Hi Xuesong, Traffic policy for subscribers is managed per PDU session at the UPF (and gNB). GTP-u does provide encapsulation between the end points, but its control fields are meant for conveying control semantics between the GTP endpoints: they were not intended for IP transport/ traffic underlays. 5QI/QCI etc are in the GTP extension header which may not be ideal to lookup to classify each packet in the transport network. The entity that classifies data packets (upstream at gNB and downstream at UPF-PSA) also inserts the DSCP for that GTP packet. The classification is based on subscriber aspects but may also on be based on its content (e.g., using DPI). Best Regards, John From: dmm <[email protected]<mailto:[email protected]>> On Behalf Of Gengxuesong (Geng Xuesong) Sent: Wednesday, January 20, 2021 8:23 PM To: Uma Chunduri <[email protected]<mailto:[email protected]>>; Lizhenbin <[email protected]<mailto:[email protected]>> Cc: [email protected]<mailto:[email protected]>; dmm <[email protected]<mailto:[email protected]>> Subject: Re: [DMM] [Apn] Regarding APN Usecase in Mobile Core Hi Uma and all, I have read the document and got a few questions: In my understanding, in the UPF where traffic policy is enforced, the fine-granularity services are provided. Then what fields in the GTP-u encapsulation indicates the traffic's service requirements? When a GTP-u tunnel goes into a SRv6 policy, according to which fields in the GTP-u encapsulation the DSCP is generated? We know that there are parameters such as 5QI/QCI and QFI, whether they are associated with a GTP-u tunnel? Best Xuesong From: Apn [mailto:[email protected]] On Behalf Of Uma Chunduri Sent: Tuesday, January 19, 2021 3:17 AM To: Lizhenbin <[email protected]<mailto:[email protected]>> Cc: [email protected]<mailto:[email protected]>; dmm <[email protected]<mailto:[email protected]>> Subject: Re: [Apn] [DMM] Regarding APN Usecase in Mobile Core Hi Robin, In-line.. Cheers! -- Uma C. On Mon, Jan 18, 2021 at 5:25 AM Lizhenbin <[email protected]<mailto:[email protected]>> wrote: Hi APNers and DMMers, I remember that in the mobile core scenarios the GTP-u tunnel can be set up according to the user and application requirements, but I do not understand the details. [Uma]: Obviously, the best reference for GTP-U is TS 29.281. However, uou should look into https://datatracker.ietf.org/doc/draft-ietf-dmm-5g-uplane-analysis/<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdatatracker.ietf.org%2Fdoc%2Fdraft-ietf-dmm-5g-uplane-analysis%2F&data=04%7C01%7Cjohn.kaippallimalil%40futurewei.com%7C8d3a8d34ea0049ed523f08d8c43ccbc9%7C0fee8ff2a3b240189c753a1d5591fedc%7C1%7C0%7C637475112703506588%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=r6m55gGEQqmacjcKy21wzksXsCzvxLBF%2Fx%2FqUFFxFPU%3D&reserved=0> where lot more details and other references related this topic was analyzed (primarily started after/during REL-15, when any other use plane other than GTP-U is worthwhile is debated for 5G N9 interface). I think when the packet tunneled by GTP-u traverses the APN-based transport network, it may be mapped to the corresponding tunnel according to the user and application requirements to implement the uniform service. If you are familiar with the principle of GTP-u in the mobile core, please help provide some details. Best Regards, Zhenbin (Robin) _______________________________________________ dmm mailing list [email protected]<mailto:[email protected]> https://www.ietf.org/mailman/listinfo/dmm<https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.ietf.org%2Fmailman%2Flistinfo%2Fdmm&data=04%7C01%7Cjohn.kaippallimalil%40futurewei.com%7C8d3a8d34ea0049ed523f08d8c43ccbc9%7C0fee8ff2a3b240189c753a1d5591fedc%7C1%7C0%7C637475112703506588%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=5lmTCot3wizMZWRXkGRQDP8c03xbCGVOqNIBo1tplx0%3D&reserved=0>
_______________________________________________ dmm mailing list [email protected] https://www.ietf.org/mailman/listinfo/dmm
