Hi Lothar,
I do not really understand what you are trying to achieve by means of your "virtual yellow cable". What the connection between a TSI and a VAP looks like depends on the kind of TS/TSI you have and where the VAP/NVE is located. It can be a physical (Ethernet) link, a logical link, or a chain of physical and logical links. It might be helpful to think about the TSI as having two "sides", but this is my personal interpretation: One side of the TSI faces the TS and its operating system, and the other side faces the network. In my opinion, it is outside the scope of NVO3 what the TS facing side looks like, because this normally depends on the operating system the TS is running. In regard to the VAP, I think of it as a physical or logical port on an NVE which can be identified by means of a physical or logical interface identifier, e.g. a physical or logical port ID, a VLAN ID and so on. Inside the NVE each VAP is associated with exactly one VNI. Thanks, Florian >>> "Reith, Lothar" <[email protected]> 22.04.13 12.03 Uhr >>> Florian, I would like to share some more thoughts regarding the virtual yellow cable, with the intend to get to a better definition and understanding of the terms “VAP” and “TSI”, and how they are connected to each other. Here is how I see the 40 years of relevant history related to the yellow cable and to NVO3: It started with a yellow cable: a physical multipoint cable. 1. Bridge: creates one virtual-yellow cable from 2 physical yellow cables to overcome distance limit 1. Hub: creates one virtual yellow cable from multiple physical yellow cables to allow structured cabling, but collisions limit throughput 2. Multiport Bridge (also called Ethernet Switch): Allows to create multiple virtual yellow cables from groups of ports terminating physical point to point cables. No more collisions but may drop frames. 3. IEEE802.1Q VLAN based Ethernet Switched networks: Allows to create upto 4K (4094) virtual yellow cables per network, one station (MAC-address) can be endpoint in upto 4K virtual yellow cables/networks/VLANs. But: Not suitable for provider as customer requires multiple VID himself. 4. IEEE Provider Bridges: “QinQ” type S-VLAN: 4K virtual yellow cables per provider independent of customer use of VLAN/virtual yellow cables 5. IEEE Backbone Provider Bridges: MAC in MAC with upto 16 Mio I-SIDs as virtual yellow cable ID. 6. VXLAN: IETF defined MAC in IP encapsulation with upto 16 Mio Segment-IDs as virtual yellow cable ID. 7. NVO3: IETF defined work in progress with the term virtual network refers either to a virtual layer2 network (virtual yellow cable), or to a virtual layer 3 network (a community of Layer 2 endpoints – i.e. interfaces to virtual yellow cables with associated layer 3 endpoint addresses). Said layer3 endpoint addresses are not used for routing, only for endpoint identification in overlay networks and associated underlay network, where routing is done based on underlay IP-addresses. Given above, it should be possible to improve the definition of TSI, by logically inverting the definition you gave: You stated (excerpt): “containment tree might look like this: TSI -- VN -- MAC -- IP.” Let me try to logically invert your definition as follows: A TSI is an interface to a virtual yellow cable connecting the TSI with a VAP, with the following granularity: 1. If an IP address is bound to a MAC address and said MAC address bound to a virtual yellow cable interface, then the TSI has the granularity of the IP. Please note that multiple IP addresses may be bound to the same MAC address. 2. If a MAC address is bound to the virtual yellow cable interface, then the TSI has the granularity of that MAC-address (like a normal station seeing only the frames destined to it and ignoring other frames) 3. the virtual yellow cable interface, then the TSI has the granularity of a VN (like a protocol analyzer looking at all frames arriving at the station or passing the station). If we are talking about a container that includes all 3 interface granularity options listed above, then we may call this a TSI as the underlying interface, which enables these options. And please excuse my terrible mistake in the emails below where I have used the word vertice as if it meant edge, even though it refers to a point in graph theory. So when you read below emails, please replace the term “vertice” with the term “edge” or “line”, in order to make sense. Von: Reith, Lothar Gesendet: Montag, 22. April 2013 10:41 An: Reith, Lothar; Florian Mahr Cc: [email protected] Betreff: AW: [nvo3] Antw: Re: NVO3 Terminology changes Sorry I screwed up the meaning of vertice. In mathematical graph theory the vertices are the points that are connected by edges. So what I meant to say is that I now know what the nature of the edge is, that connects the both vertices “VAP” and “TSI”. It is a collapse virtual yellow cable. Apologies for the confusion. Lothar Von: Reith, Lothar Gesendet: Montag, 22. April 2013 10:36 An: Reith, Lothar; Florian Mahr Cc: [email protected] Betreff: AW: [nvo3] Antw: Re: NVO3 Terminology changes Heureka, I now know what the vertice is. In the case where the TSI is the interface of a VM, it is a collapsed virtual yellow cable, the thing that Bob Metcalfe called “the Ether” in his original 40 year old drawing. And both the VAP as well as the TSI are Interfaces to the “Ether”. And they are both of point nature.s Is there any other case in scope of NVO3, where the TSI is not an interface of a VM? The current definition of Tenant System may have mislead me to believe so. Lothar Von: [email protected] [mailto:[email protected]] Im Auftrag von Reith, Lothar Gesendet: Montag, 22. April 2013 10:07 An: Florian Mahr Cc: [email protected] Betreff: Re: [nvo3] Antw: Re: NVO3 Terminology changes Hi Florian, thanks for this interpretation. I now understand that I may indeed have been confused, and that the TSI is not b). I am still not quite clear, if it’s a) or c) though. For me, that is a very fundamental conceptual difference – like for Euklid the difference between a point and a line that looks like a point, because it is a collapsed line. My take is, that you tend for a) – correct? That means that the TSI corresponds to the VAP in a 1 to 1 relation (because they are 2 points connected to each other by said implicit point to point relation, which represents the vertice in the graph, however, the 2 points are the endpoints of a collapsed line, and the line (the vertice between the point “VAP” and the point “TSI” is not present in the data model. Is that what you are saying? Lothar Von: Florian Mahr [mailto:[email protected]] Gesendet: Sonntag, 21. April 2013 20:34 An: Reith, Lothar Cc: [email protected] Betreff: Antw: Re: [nvo3] NVO3 Terminology changes Hi Lothar, I agree with you that the discussion that is going on about the term TSI is a little bit confusing. In my opinion, it is best to think about the TSI in regard to the containment tree proposed in draft-kreeger-nvo3-hypervisor-nve-cp-01. In its simplest case the containment tree might look like this: TSI -- VN -- MAC -- IP. Therefore, a TSI is nothing more than a conceptual interface which connects a Tenant System (TS) to a Virtual Network Instance (VNI). I personally like the new term TSI more than the old one (VNIC), because it does not make any assumption about the implementation of the interface, i.e. the TSI can be for example a logical NIC or one-to-one relationship between the TSI on the TS side and the VAP on the NVE side, i.e. each TSI is logically associated with exactly one VAP. I do not think it is a good idea to call the TSI an "Overlay network interface", because the TSI is not directly involved in creating the overlay network. This is the job of the NVE which conceptually consists of one or more VNIs and the Overlay Module. Thanks, Florian >>> "Reith, Lothar" <[email protected]> 21.04.13 13.44 Uhr >>> Given that the term TSI lead to a long discussion and perhaps even confusion, I like to offer some abstraction help. Please consider any network can be drawn as a graph with vertices and points representing the endpoints of the vertices. Agreed? If you agree, please consider which of the 3 options the TSI represents: a) The point in the tenant system that interfaces towards the network b) The point in the network that interfaces towards the tenant system c) The vertice that connects 2 points (one in the network that is closest to the tenant system and one in the tenant system that is closest to the network) Do all agree that it is b) ? If yes, do you agree that we may be better of in renaming Tenant System Interface as “Overlay network interface”? Or did I get something totally wrong? Lothar Von: [email protected] [mailto:[email protected]] Im Auftrag von Lizhong Jin Gesendet: Freitag, 19. April 2013 19:05 An: [email protected]; Lawrence Kreeger Betreff: Re: [nvo3] NVO3 Terminology changes Hi Larry, I wonder if point C or D is feasible in implementation. In my understanding, the TSI must be a virtual port which could be associated with the special VN, and transmit/receive packets from tenant system (the virtual port terminology is used in many switch chip). If we use one IP address, we could not treat it as a virtual port. One simple example, if packet with broadcast IP address received, will this packet be send to IP_a and IP_b on the same vNIC? I don't think we need to define such kind of detail for TSI. The TSI definition is implementation related. I could define the TSI with a VLAN (piont B in your figure), or define TSI with a priority queue within one vNIC. BTW, I like the name TSI, and remove NIC/vNIC/pNIC. pNIC is closely related with data transfer between network and host stack, and it is usually a controller without MAC&PHY. vNIC is a driver function or virtual function(defined by SR-IOV). A TSI could also be a vNIC teaming interface by teaming driver. So "interface" is a appropriate name to define a Tx/Rx channel between tenant system and VN. Regards Lizhong ---------- Forwarded message ---------- From: "Larry Kreeger (kreeger)" <[email protected]> To: Qin Wu <[email protected]>, Anoop Ghanwani <[email protected]>, "Black, David" <[email protected]> Cc: "[email protected]" <[email protected]>, "Reith, Lothar" <[email protected]> Date: Wed, 17 Apr 2013 16:55:50 +0000 Subject: Re: [nvo3] NVO3 Terminology changes Hi Qin, A VM is one possible implementation of a Tenant System. I have not seen anyone suggesting that a Tenant System get more granular than that. Regarding what defines a TSI, someone could choose any branch in the tree below to define as the TSI. I was assuming that the TSI would be at point A below in the tree. However, point B would correspond the the VAP. Pat had mentioned having the TSI at point D. I think if we were to define the TSI to be either point C or point D, then it would depend on the service being given (L2 or L3). I would prefer it to be at point A or B. VNIC-+-VN-+-MAC-+-IP ^ | | +-IP ... | | | | | +-MAC-+-IP | | +-IP ... | | | +-VN-+-MAC-+-IP - Larry
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
