Hi Pablo, Please see zzh2 > below.
-----Original Message----- From: Pablo Camarillo (pcamaril) <[email protected]> Sent: Friday, May 14, 2021 4:15 AM To: Jeffrey (Zhaohui) Zhang <[email protected]>; [email protected] Subject: RE: [DMM] [spring] note: WGLC on draft-ietf-dmm-srv6-mobile-uplane-11 [External Email. Be cautious of content] Hi Jeffrey, Many thanks for your reply. Good to see some of your points are resolved now. Answers to the remaining points below inline with [PC2]. Cheers, Pablo. -----Original Message----- From: Jeffrey (Zhaohui) Zhang <[email protected]> Sent: miércoles, 5 de mayo de 2021 20:37 To: Pablo Camarillo (pcamaril) <[email protected]>; [email protected] Cc: '[email protected]' <[email protected]> Subject: RE: [DMM] [spring] note: WGLC on draft-ietf-dmm-srv6-mobile-uplane-11 Hi Pablo, A lot of my questions/confusion came from the fact that in many cases described in this document, the gNB/UPF will use SRv6 in place of GTP-U based on changed N2/N4. That was not clear to me because my understanding is that 3GPP has *not* agreed on using SRv6 in place of GTP-U. [PC2] As stated in the draft "The user plane described in this document does not depend on any specific architecture." Zzh2> Because many sections are based on changed N2/N4 interfaces, so this document does depend on a changed architecture not yet approved by 3GPP, with the exception of drop-in mode (which is transparent to 3GPP)? I could be wrong on that, but if the understanding is correct, I do not think IETF/DMM should standardize this. At most an experimental document to show how IETF views SRv6 could be used for mobile user plane. The drop-in mode is fine, but I am curious why bother have GWs messing with the packets, and what are the security implications. I still need some help understanding the modes better. Please see zzh> below. I snipped some exchanges to focus remaining questions/comments. -----Original Message----- From: Pablo Camarillo (pcamaril) <[email protected]> Sent: Tuesday, May 4, 2021 6:36 AM To: Jeffrey (Zhaohui) Zhang <[email protected]>; [email protected] Subject: RE: [DMM] [spring] note: WGLC on draft-ietf-dmm-srv6-mobile-uplane-11 [External Email. Be cautious of content] Hi Jeffrey, Many thanks for your review. We've posted an updated revision of the document (rev12). Comments inline with [PC]. Thanks, Pablo. -----Original Message----- From: dmm <[email protected]> On Behalf Of Jeffrey (Zhaohui) Zhang Sent: miércoles, 21 de abril de 2021 6:49 To: Erik Kline <[email protected]>; SPRING WG <[email protected]>; [email protected] Subject: Re: [DMM] [spring] note: WGLC on draft-ietf-dmm-srv6-mobile-uplane-11 Hi, I noticed this late and just did a rushed review of the document, so pardon if my questions and comments do not make sense. It seems that there is no real difference between traditional mode and enhanced mode - it's just whether an SRH is used for TE purpose or not. Is there a need to define these two modes? For example, section "5.3. Enhanced mode with unchanged gNB GTP behavior" should work with traditional mode as well. 5.2 does talk about "scalability" and "service programming", but the they're not elaborated by the example. I think it's important to elaborate since you introduced the two modes. [PC] The example does talk about service programming (S1 is a VNF). Ack on the scalability, which is briefly mentioned but not explained in detail. I've added more on it to the document. Thanks. Zzh> It seems to me that "service programming" and TE are both transparent; even for the "traditional" mode, the two functions can be applicable. [PC2] It is different for the source node, who needs to encode a SID list into the packet. Zzh2> Section 5.1 only talks about "there will be a unique SRv6 SID associated with each PDU Session" but does not talk SID list. If SID list is a key difference, it should be mentioned in 5.1 as well. Zzh2> Section 5.2 says: The gNB MAY resolve the IP address received via the control plane into a SID list using a mechanism like PCEP, DNS-lookup, LISP control-plane or others. The resolution mechanism is out of the scope of this document. Zzh2> I don't understand why the above can't be done for 5.1 as well. That's why I was asking why the SID list is a difference between the two modes. In fact, I see the following later: Even though we have presented these methods as an extension to the "Enhanced mode", it is straightforward in its applicability to the "Traditional mode". So I wonder why you need to define two modes. [PC] Traditional mode is simply a replacement of GTP-U as overlay protocol by SRv6. This minimizes changes to the mobile architecture. The enhanced mode allows to have intermediate SIDs for TE and NFV, but also allows to aggregate several devices under the same SID -which improves scalability-. The differences are covered in Section 5.2. Section 5.3 presents interworking mechanisms for the case whereas the N3 interface is unmodified. The interworking mechanisms presented in 5.3 are applicable to both Traditional and Enhanced mode. Zzh> About "This minimizes changes to the mobile architecture", I was assuming that it means the N2/N4 interface does not change and traditional <tunnel endpoint address, TEID> tuple is still used in the control plane to identify a PDU session, but the gNB/UPF will map between the tuple (received on N2/N4) and SRv6 SID. Now it seems that N2/N4 are changed and GTP is not used at all. For that I don't think "this minimizes the changes to the mobile architecture". [PC2] On the control plane you send an IPv6 address, which now will be an SRv6 SID. In enhanced mode you need to pass a SID list of more than one SID, which needs modification on N2/N4 to be able to support it. Zzh2> For traditional mode don't you also need to modify N2/N4 to instruct gNB/UPF to use SRv6 instead of GTP-U? Zzh2> In fact, your above text seems to suggest that the SID list is passed in N2/N4, which is contradicting to the following: The gNB MAY resolve the IP address received via the control plane into a SID list using a mechanism like PCEP, DNS-lookup, LISP control-plane or others. The resolution mechanism is out of the scope of this document. Zzh2> Can you clarify how the SID list is installed on gNB? Zzh> On the other hand, the "drop-in" mode described later in 5.4 really "minimizes changes to the mobile architecture". Additionally, whatever mode it is (traditional/drop-in, enhanced w/ or w/o interworking), intermediate SIDs for TE/NFV should be applicable. [PC2] Drop-in mode does not have any change on the mobile architecture. Indeed the mobile architecture is completely agnostic from the SRv6 dataplane. Zzh2> That's what I mean, and that's the only mode I think that IETF/DMM can specify. Though some details are missing, e.g. how are all GTP-U functions achieved via SRv6 - there are other fields in the GTP header besides the tunnel destination address and TEID. Zzh> Section 5 says: In order to simplify the adoption of SRv6, we present two different "modes" that vary with respect to the use of SRv6. The first one is the "Traditional mode", which inherits the current 3GPP mobile user- plane. In this mode GTP-U protocol [TS.29281] is replaced by SRv6, however the N3, N9 and N6 interfaces are still point-to-point interfaces with no intermediate waypoints as in the current mobile network architecture. Zzh> Don't " inherits the current 3GPP mobile user-plane" and "GTP-U protocol [TS.29281] is replaced by SRv6" contradict each other? The "inherits ..." caused major confusion to me and it should be reworded. [PC2] There's a typo in the first sentence. s/user-plane/architecture. Updated draft. Zzh> My understanding is that N3/N9 interfaces are logical. Whether there are intermediate TE/NFV waypoints is transparent to 5GS (unless the NVF are actually 5G functions), so I don't understand why that is a distinguisher between the traditional and enhanced mode. Zzh> I understand that a gNB may not be able to push SRHs, but that should not become a factor of categorizing modes. Scalability is a reasonable factor but more on that later. [PC2] The difference is on the control plane towards the gNB (or its local policy in absence of a control plane), not a question of dataplane capabilities. Zzh2> If it's based on local policy, or if the control plane is not N2/N4, then the use of SRH does not seem to be a real difference between the two modes. Why can't it be applied to "traditional" mode as well? Zzh> 5.2 says: The gNB control-plane (N2 interface) is unchanged, specifically a single IPv6 address is provided to the gNB. Zzh> For clarity/parity, 5.1 should explicitly say that N2/N4 interfaces are changed. [PC2] Added in 5.1 Zzh2> Now the above quoted text for 5.2 is a bit misleading. Obviously, the N2/N4 interface is changed in 5.2 as well, because it is now instructing gNB/UPF to use SRv6 instead of GTP-U? Zzh> 5.2.1 says: UE sends its packet (A,Z) on a specific bearer to its gNB. gNB's control plane associates that session from the UE(A) with the IPv6 address B. gNB's control plane does a lookup on B to find the related SID list <S1, C1, U2::1>. Zzh> What is that "IP address B"? The endpoint address of the tunnel? What's the TEID to use, and where is that TEID encoded in the packet (since we no longer has per-session SID)? [PC2] B is an address passes through control plane, that is used in the gNB to resolve the SID list. There is no TEID per-se, but instead different SIDs identify different sessions. Zzh2> So N2/N4 interface is indeed changed, not that "The gNB control-plane (N2 interface) is unchanged"? Zzh> 5.2.2 says: When the packet arrives at the UPF2, the UPF2 maps that particular flow into a UE PDU Session. This UE PDU Session is associated with the policy <C1, S1, gNB>. The UPF2 performs a H.Encaps.Red operation, encapsulating the packet into a new IPv6 header with its corresponding SRH. Once the packet arrives at the gNB, the IPv6 DA corresponds to an End.DX4, End.DX6 or End.DX2 behavior at the gNB (depending on the underlying traffic). The gNB decapsulates the packet, removing the IPv6 header and forwards the traffic toward the UE. Zzh> While UPF2 can identify the PDU session based on the UE address, how is that session information encoded in the packet to gNB? Policy <C1, S1, gNB> is not session specific. I assume gNB is not a simple address, but a prefix followed by the arg bits that represents the session information. It should be made clear here. [PC2] I believe this is stated in 5.2.2 Once the packet arrives at the gNB, the IPv6 DA corresponds to an End.DX4, End.DX6 or End.DX2 behavior at the gNB (depending on the underlying traffic). The gNB decapsulates the packet, removing the IPv6 header and forwards the traffic toward the UE. Zzh2> So is this (downlink) the same as in traditional mode? Zzh> RFC8986 says: The End.DX4 SID MUST be the last segment in an SR Policy, and it is associated with one or more L3 IPv4 adjacencies J. The End.DX6 SID MUST be the last segment in an SR Policy, and it is associated with one or more L3 IPv6 adjacencies J. zzh> As I understand it, the connection between gNB and UE is *not* an IP/Ether adjacency, so not sure if DX2/4/6 can be used. [PC2] Can you please clarify? If the PDU is IP, then the adjacency is IP. Regardless of whether for example you send this traffic using pdcp. Zzh2> For IP PDU, the IP adjacency is between the UPF and UE, not between gNB and UE. My understanding of DX2/4/6 is that the cross-connect is on a local L2/IP PE-CE adjacency. Zzh> 5.2.3 says: The Enhanced Mode improves since it allows the aggregation of several UEs under the same SID list. For example, in the case of stationary residential meters that are connected to the same cell, all such devices can share the same SID list. This improves scalability compared to Traditional Mode (unique SID per UE) and compared to GTP-U (dedicated TEID per UE). Zzh> The above compares it to both "unique SID per UE" and "dedicated TEID per UE". It means that the N2/N4 interface does not distinguish session anymore (at least for uplink traffic) and the aggregation is provided by the 5G itself and would work for GTP-U as well, and not provided by SRv6 or by the enhanced mode. [PC2] Neither traditional mode, nor GTP-U allows sharing TEIDs for different UEs. Hence the scale improvement. Zzh2> My point is that the sharing is not because of SRv6, but because there are cases you where you don't need to distinguish the source UE. That, if adopted by 3GPP, can be easily done to GTP-U as well. Zzh> Per earlier questions/comments, the downlink traffic still has session information in the SRv6 header (so that END.DX is done), so the same SID list can't be used, and the scalability improvement is only for uplink traffic? [PC2] As stated in the beginning of 5.2 Additionally in this mode the operator may choose to aggregate several devices under the same SID list (e.g., stationary residential meters connected to the same cell) to improve scalability. Zzh2> Right - I wanted to confirm that the aggregation is only for uplink traffic. Is that so? 5.3.1 says: o The gNB is unchanged (control-plane or user-plane) and encapsulates into GTP (N3 interface is not modified). o The 5G Control-Plane (N2 interface) is unmodified; one IPv6 address is needed (i.e. a BSID at the SRGW). It seems that N4 needs to be changed so that SMF can tell UPF1 and UPF2 to use SRv6 instead of GTP. Or is that N4 still assuming GTP but UPF1 and UPF2 are configured to turn to GTP parameters into SRv6 stuff? That should be made clear here. [PC] The UPFs are SRv6 aware (and thus N4 interface as well). The gNB is the legacy device that still runs GTPU. This is explained at the beginning of the section. Zzh> UPFs being SRv6 aware does not necessarily imply N4 interface will actually tell UPFs to use SRv6 tunnel instead of GTPU. My impression is that 3GPP has not approved to use SRv6 yet so please make it clear if this document assumes that 3GPP approves using SRv6 via changed N2/N4 (vs. the devices autonomously use SRv6 based on signaled GTP-U parameters). [PC2] This document does not "assume that 3GPP approves using SRv6 via changed N2/N4" Zzh2> If 3GPP does not approve that, then only the drop-in mode can work? Zzh> Besides, it is strange that 5GS will tell gNB to use GTP while telling UPF to use SRv6. [PC2] The decision to use of SRv6 is a local policy at the UPF. Alternatively, you could also choose to instruct this through a new control plane, but this is out of the scope of this document. Zzh2> If so, please make it clear that if N4 does not change, then the UPF make decision to use SRv6 per local policy, based on GTP-U parameters signaled on N4. That's why I was asking "Or is that N4 still assuming GTP but UPF1 and UPF2 are configured to turn to GTP parameters into SRv6 stuff? That should be made clear here"? In the following: 5.3.2.2. Packet flow - Downlink The downlink packet flow is as follows: UPF2_in : (Z,A) -> UPF2 maps flow with SID <C1, S1,SRGW::SA:DA:TEID> UPF2_out: (U2::1, C1)(SRGW::SA:DA:TEID, S1; SL=2)(Z,A) ->H.Encaps.Red C1_out : (U2::1, S1)(SRGW::SA:DA:TEID, S1; SL=1)(Z,A) S1_out : (U2::1, SRGW::SA:DA:TEID)(Z,A) SRGW_out: (SA, DA)(GTP: TEID=T)(Z,A) -> End.M.GTP4.E gNB_out : (Z,A) How does UPF2 know the address of the gNB? In IPv6 case (5.3.1) the UPF1 knows the gNB address from the N4 signaling but shouldn't IPv4 case be the same? [PC] In both 5.3.1 and 5.3.2 the gNB address is provided in the SRH. In 5.3.1 it is provided as the next SID; in 5.3.2 it is provided as SID arguments. Zzh> For UPF2 to put the gNB address into the SRH, it has to learn it first. How/where does UPF2 learn the gNB address (for either IPv4 or IPv6 case)? My understanding is that when there is a UPF1 in place, UPF2 only knows about UPF1 not gNB. [PC2] Your understanding is correct, and since in this case there is no intermediate UPF, UPF2 only knows about gNB. [PC2] If there would be an I-UPF, then UPF2 would not know about gNB, and it would use an End.MAP SID at UPF1. Zzh2> Apparently, in Figure 6, there is a UPF1? Zzh> Additionally, since you mention SRGW is separate from UPF1, using the "SRGW::" prefix is confusing. How is UPF2 supposed to know about the SRGW which is supposed/assumed to be transparent? [PC2] Local policy at the UPF2. For the following: 5.3.2.3. Scalability Is it the same as IPv6 case? If so it's better to just state so instead of repeating. [PC] It is not exactly the same, as in IPv6 the use of the BSID yields remote steering; while in the IPv4 case you need to have an Uplink Classifier in the SRGW that performs packet classification. While the diff is small, it is not equivalent. Zzh> I need to think this through. May come back to this later. What is a L3-anchor vs. L2-anchor? [PC] Different types of 3GPP UPF functionalities. Equivalent to SGW/PGW in 4G. Zzh> Can you give a reference for my better understanding? I can't find it. [PC2]. Sure 4.1.4.2 of https://urldefense.com/v3/__https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=728__;!!NEt6yMaO-gk!RWgox3-e0-NCn506Dily-LC6qemqom36PnLRe0zsQOp6wg1L8TshqU0jmz-4A2GJ$ Zzh2> 4.1.4.2 of 23.002 does not talk about l3/l2-anchor. Can you use well-known terms from 5G specifications? For 6.4, how is the SGB::TEID written into the SRH? If it's via "S04. Push a new IPv6 header with its own SRH containing B", does that mean we need one policy for each TEID? [PC] It is pushed in line S08. Zzh> The question is how the GTP information (e.g., TEID) is put into in the SRv6 header. Apparently, the policy needs to extract the TEID from the GTP header but the instructions does not talk about (S08 is about S which does not have the information). [PC2] Sure, we can make that explicit in S08. For the following rules in 6.5: S01. If (Next Header = UDP & UDP_Dest_port = GTP) { S02. Copy SRH[0] to buffer memory ... S08. Set the GTP TEID (from buffer memory) Does SRH[0] corresponds to U::1 or SGB::TEID in the 5.4 example as quoted below? C1_out : (SRGW-A, SGB::TEID)(U::1, SGB::TEID, C1; SL=1)(A,Z) GW-B_out: (SRGW-B, U::1)(GTP: TEID T)(A,Z) ->U1b::TEID is an End.M.GTP6.E SID at SRGW-B [PC] U::1 Zzh> Since U::1 does not have TEID, how does S08 get TEID "from the buffer memory"? [PC2] As stated in the Note right after the pseudocode: " The TEID is extracted from the argument space of the current SID." Zzh2> To do the extraction, I expected to see "Copy S to buffer memory" at step S02, but it's not there, hence the question. Zzh2> Additional question - I had thought that the UPF is using SRv6 w/o GTP. So when downlink traffic gets to the SRGW in 5.3.1.2, how is the S01 condition met (Next Header = UDP & UDP_Dest_port = GTP)? Does the SRv6 packet carry a complete UDP/GTP header (I had thought there is no UDP/GTP header at all when traffic leaves the UPF because we're using SRv6)? If yes, how is that different from transporting GTP over SRv6? Zzh> A new question about 5.4: gNB_out : (gNB, U::1)(GTP: TEID T)(A,Z) GW-A_out: (GW-A, S1)(U::1, SGB::TEID, C1; SL=3)(A,Z)->U::1 is an End.M.GTP6.D.Di SID at SRGW-A S1_out : (GW-A, C1)(U::1, SGB::TEID, C1; SL=2)(A,Z) C1_out : (GW-A, SGB::TEID)(U::1, SGB::TEID, C1; SL=1)(A,Z) GW-B_out: (GW-B, U::1)(GTP: TEID T)(A,Z) ->SGB::TEID is an End.M.GTP6.E SID at SRGW-B Zzh> gNB sends (gNB, U::1) packets and UPF also expects the same gNB source address. However, GW-B sends (GW-B, U::1) - wouldn't this cause trouble on the UPF if the there are security checks for the validity of source addresses? [PC2] The UPF has a route to the gateway, and hence uRPF does not fail. If this is what you mean... Zzh2> This is not about uRPF. I assume UPF would not just accept tunnel packets from any source address (to which it has a route) - it would only accept from know addresses it learned from SMF, right? Zzh2> Thanks. Zzh2> Jeffrey Zzh> Thanks. Zzh> Jeffrey Thanks. Jeffrey -----Original Message----- From: spring <[email protected]> On Behalf Of Erik Kline Sent: Wednesday, April 7, 2021 11:59 PM To: SPRING WG <[email protected]> Cc: [email protected]; [email protected]; [email protected] Subject: [spring] note: [DMM] WGLC on draft-ietf-dmm-srv6-mobile-uplane-11 [External Email. Be cautious of content] Spring folk, Per request from spring chairs, I wanted to drop a note that a document discussing using SRv6 in a 3GPP carrier network is in dmm WGLC. If this is of interest to you, please feel free to read and review (please direct comments to dmm@). Thanks, -Erik On Wed, Apr 7, 2021 at 10:35 AM Sri Gundavelli (sgundave) <[email protected]> wrote: > > Working Group: > > > > As we discussed in the last IETF meeting, we are issuing WGLC on > draft-ietf-dmm-srv6-mobile-uplane-11. > > The document went through several revisions and there were good amount of > reviews on this document. I am very pleased with the quality of this > document. The authors have addressed all the comments and there are no open > issues that we are tracking at this time. We believe the document is ready > for IESG reviews and like to confirm the same from the working group. > > > > The following message commences a two week WGLC for all feedback. > > > > Document Link: > > https://urldefense.com/v3/__https://www.ietf.org/archive/id/draft-ietf > -dmm-srv6-mobile-uplane-11.txt__;!!NEt6yMaO-gk!RyYI05Z41Depmow2gAcvg-R > TNNFjsP2hvLNFyz-fGE_jQjZlyvBbW-zDgRDc4fJS$ > > > > > > Please post any comments/concerns on this document. > > > > > > Thanks! > > Sri > > > > > > > > _______________________________________________ > dmm mailing list > [email protected] > https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/dmm_ > _;!!NEt6yMaO-gk!RyYI05Z41Depmow2gAcvg-RTNNFjsP2hvLNFyz-fGE_jQjZlyvBbW- > zDgY0wXO9D$ _______________________________________________ spring mailing list [email protected] https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/spring__;!!NEt6yMaO-gk!RyYI05Z41Depmow2gAcvg-RTNNFjsP2hvLNFyz-fGE_jQjZlyvBbW-zDgQZXTemu$ Juniper Business Use Only _______________________________________________ dmm mailing list [email protected] https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/dmm__;!!NEt6yMaO-gk!VfY3TV_tyoDu-Jl2ZATcisDxW0ThG3_0UhTy33UgBAM_eiwNmnAps4fqtSRX3Ayz$ Juniper Business Use Only Juniper Business Use Only Juniper Business Use Only Juniper Business Use Only _______________________________________________ dmm mailing list [email protected] https://www.ietf.org/mailman/listinfo/dmm
