Thanks for the input Wido, > That said, you could also opt that you can specify BGP peers are zone level > and override them at network level if one prefers. Nothing specified at >the > network? The zone-level peers are used. If you do >specify them at the > network level those are used. Again, think about multihop.
That's exactly what I had in mind, the same way we set DNS for the zone but can specify at network level as well. This way we keep self-service intact for when end-users simply want a routed network that peers with whatever the provider has setup upstream but also give the ability to either peer with an user managed VNF upstream or an operator provided router. I hope this way we can cater for most use cases, at least with a first simple implementation. Will definitely keep multihop in mind. Cheers, Alex -----Original Message----- From: Wido den Hollander <w...@widodh.nl> Sent: Monday, May 20, 2024 8:22 PM To: d...@cloudstack.apache.org; Alex Mattioli <alex.matti...@shapeblue.com>; users@cloudstack.apache.org; adietr...@ussignal.com Subject: Re: Dynamic routing for routed mode IPv6 and IPv4 Isolated and VPC networks Op 20/05/2024 om 14:45 schreef Alex Mattioli: > Hi Alex, > > In this scenario: > >> I think adding the ability to add network specific peers as mentioned in one >> of >your prior replies would still allow the level of control some operators >> (myself >included) may desire. > > How do you propose network specific peers to be implemented? > I do agree with Alex (Dietrich) that I think BGP peers should be configured per network. There is no guarantee that every VLAN/VNI (VXLAN) ends up at the same pair of routers. Technically there is also no need to do so. Let's say I have two VNI (VXLAN): VNI 500: Router 1: 192.168.153.1 / 2001:db8::153:1 Router 2: 192.168.153.2 / 2001:db8::153:2 VNI 600: Router 1: 192.168.155.1 / 2001:db8::155:1 Router 2: 192.168.155.2 / 2001:db8::155:2 In these case you would say that the upstream BGP peers are .153.1/2, .155.1/2 (and their IPv6 addresses). No need for BGP multihop. Talking about multihop, I would make that optional, people might want to have two central BGP routers where each VR peers with (multihop) and those routers distribute the routes into the network again. Per network you create you also provide the ASN range, but even better would be to refer to a pool. You can use one pool for your zone by referencing every network to the same pool are simply use multiple pools if your network requires so. That said, you could also opt that you can specify BGP peers are peer level and override them at network level if one prefers. Nothing specified at the network? The zone-level peers are used. If you do specify them at the network level those are used. Again, think about multihop. Wido > Regards > Alex > > > > > > -----Original Message----- > From: Dietrich, Alex <adietr...@ussignal.com.INVALID> > Sent: Monday, May 20, 2024 2:21 PM > To: users@cloudstack.apache.org; d...@cloudstack.apache.org > Subject: Re: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks > > Hi Alex, > > This may be a difference in perspective in implementation of BGP at the > tenant level. I see the ability this would provide to seamlessly establishing > those peering relationships with minimal intervention (helping scalability). > > I think adding the ability to add network specific peers as mentioned in one > of your prior replies would still allow the level of control some operators > (myself included) may desire. > > Thanks, > Alex > > [photo]<http://www.ussignal.com/> > > Alex Dietrich > Senior Network Engineer, US Signal > > 616-233-5094<tel:616-233-5094> | > www.ussignal.com<https://www.ussignal.com> | > adietr...@ussignal.com<mailto:adietr...@ussignal.com> > > 201 Ionia Ave SW, Grand Rapids, MI > 49503<https://maps.google.com/?q=201%20Ionia%20Ave%20SW,%20Grand%20Rap > ids,%20MI%2049503> > > [linkedin]<https://www.linkedin.com/company/us-signal/> > > [facebook]<https://www.facebook.com/ussignalcom/> > > [youtube]<https://www.youtube.com/channel/UCaFBGFfXmHziWGTFqjGzaWw> > > IMPORTANT: The contents of this email are confidential. Information is > intended for the named recipient(s) only. If you have received this email by > mistake, please notify the sender immediately and do not disclose the > contents to anyone or make copies thereof. > > [__tpx__] > From: Alex Mattioli <alex.matti...@shapeblue.com> > Date: Monday, May 20, 2024 at 7:51 AM > To: users@cloudstack.apache.org <users@cloudstack.apache.org>, > d...@cloudstack.apache.org <d...@cloudstack.apache.org> > Subject: RE: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks EXTERNAL > > Hi Alex, > >> I am not convinced that specifying BGP peers at the zone level is a good >> idea given the impacts BGP can have on a given network. I would much rather >> see both peer and AS specification handled at the >network configuration, or >> another more specific level. > > I don't see how else end users would be able to automatically create routed > networks without intervention from the operator. > > > Cheers > Alex > > > > > -----Original Message----- > From: Dietrich, Alex <adietr...@ussignal.com.INVALID> > Sent: Thursday, May 16, 2024 2:23 PM > To: users@cloudstack.apache.org; d...@cloudstack.apache.org > Subject: Re: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks > > Hello Alex, > > I appreciate this back and forth as I am excited about the potential this > feature would hold. > > > * This is a very valid point. We could add network specific BGP peers > as well, which would override the automatic AS allocation, in the same way > that we now allocate DNS servers in the zone level but can override that by > manually selecting different DNS servers at network creation time. Would > that address your point? > > Why does the network specific BGP peers need to override automatic AS > allocation? In my mind there isn’t a dependency that needs to exist to those > two as they are somewhat independent of one another. > > I am not convinced that specifying BGP peers at the zone level is a good idea > given the impacts BGP can have on a given network. I would much rather see > both peer and AS specification handled at the network configuration, or > another more specific level. > > Thanks, > Alex > > From: Alex Mattioli <alex.matti...@shapeblue.com> > Date: Wednesday, May 15, 2024 at 10:15 AM > To: users@cloudstack.apache.org <users@cloudstack.apache.org>, > d...@cloudstack.apache.org <d...@cloudstack.apache.org> > Subject: RE: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks EXTERNAL > > Hi Alex, > >> Would zone-level BGP peers be those used by default for establishing new BGP >> peers in networks where dynamic routing is enabled? > > Correct, so far we plan to allow for up to 4 BGP peers for a zone, with the > possibility to setup different metrics to each peer. > >> This could affect a multi-tenant model where there may be different BGP >> peers presented based on what the upstream network provides. An example of >> >this would be where the VLANs associated to a given account are associated >> to distinct VRFs and may have different peering IP addresses. >> I would like to see the peering IP addresses specific to the networks where >> dynamic routing is enabled instead of specifying defaults at the zone level. > > > This is a very valid point. We could add network specific BGP peers as well, > which would override the automatic AS allocation, in the same way that we now > allocate DNS servers in the zone level but can override that by manually > selecting different DNS servers at network creation time. Would that address > your point? > > Cheers, > Alex > > > > > -----Original Message----- > From: Dietrich, Alex <adietr...@ussignal.com.INVALID> > Sent: Wednesday, May 15, 2024 2:34 PM > To: users@cloudstack.apache.org; d...@cloudstack.apache.org > Subject: Re: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks > > Hi Alex, > > I appreciate the clarity! > > Excuse my ignorance if I am misunderstanding the intention of specifying BGP > peers at the zone level. > > Would zone-level BGP peers be those used by default for establishing new BGP > peers in networks where dynamic routing is enabled? > > This could affect a multi-tenant model where there may be different BGP peers > presented based on what the upstream network provides. An example of this > would be where the VLANs associated to a given account are associated to > distinct VRFs and may have different peering IP addresses. > > I would like to see the peering IP addresses specific to the networks where > dynamic routing is enabled instead of specifying defaults at the zone level. > > > * Alex > > [__tpx__] > From: Alex Mattioli <alex.matti...@shapeblue.com> > Date: Wednesday, May 15, 2024 at 9:27 AM > To: users@cloudstack.apache.org <users@cloudstack.apache.org>, > d...@cloudstack.apache.org <d...@cloudstack.apache.org> > Subject: RE: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks EXTERNAL > > Hi Alex, > > Answers inline below with > > > Cheers > > > > > -----Original Message----- > From: Dietrich, Alex <adietr...@ussignal.com.INVALID> > Sent: Wednesday, May 15, 2024 3:12 PM > To: users@cloudstack.apache.org; d...@cloudstack.apache.org > Subject: Re: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks > > Hello Alex, > > I appreciate you taking on this initiative as I’d like to see similar > functionality made available in CloudStack. > > I do have some feedback on your implementation approach: > > 1 - Operator configures one or more BGP peers for a given Zone (with > different metrics) > > What is the intention behind specifying BGP peers at the zone level? I would > think this would need to be specific to the network that you want to enable > BGP on and does not need to concern the entire zone. > >> The goal is for the process to be drive by the end user without operator >> intervention. In the current design we'd enable the VR to share routes with >> upstream routers without any need for extra configuration on the part of the >> operator. >> Your point is very valid and it should definitely be a future enhancement on >> the feature. > > 2 - Operator presents a pool of Private AS numbers to the Zone (just > like we do for VLANs) > > As a private AS consumer, I agree that this approach would be helpful for a > more dynamic allocation as new dynamic routing enabled networks are created. > >> Glad we are in the same page there. > > 3 - When a network is created with an offering which has dynamic > routing enabled an AS number is allocated to the network > > 4 - ACS configures the BGP session on the VR (using FRR), advertising > all its connected networks > > Given there is a lot of extensibility within BGP, I would think there would > need to be some level of customizability to the peering configurations. Is > the intention to consider adding additional knobs, or relegating that to the > upstream BGP peer? I could see scenarios where you would at least want to > have control over prefix lengths, etc. > >> Absolutely, but I think this should be a future enhancement, the current >> goal is to have a very simple and basic dynamic BGP implementation working, >> after that's out there and in use then we definitely should discuss how to >> enhance the >feature with exactly what you pointed out. > > > Thanks, > Alex Dietrich > > > [__tpx__] > From: Alex Mattioli <alex.matti...@shapeblue.com> > Date: Wednesday, May 15, 2024 at 8:55 AM > To: users@cloudstack.apache.org <users@cloudstack.apache.org>, > d...@cloudstack.apache.org <d...@cloudstack.apache.org> > Subject: RE: Dynamic routing for routed mode IPv6 and IPv4 Isolated > and VPC networks EXTERNAL > > Hi all, > > Does anyone have an opinion on the implementation of dynamic routing in > Isolated networks and VPCs? > > So far the design is: > > 1 - Operator configures one or more BGP peers for a given Zone (with > different metrics) > 2 - Operator presents a pool of Private AS numbers to the Zone (just > like we do for VLANs) > 3 - When a network is created with an offering which has dynamic > routing enabled an AS number is allocated to the network > 4 - ACS configures the BGP session on the VR (using FRR), advertising > all its connected networks > > Any and all input will be very welcome. > > Cheers, > Alex > > > > > From: Alex Mattioli > Sent: Wednesday, April 17, 2024 3:25 AM > To: users@cloudstack.apache.org; d...@cloudstack.apache.org > Subject: Dynamic routing for routed mode IPv6 and IPv4 Isolated and > VPC networks > > Hi all, > > I'd like to brainstorm dynamic routing in ACS (yes, again... for the > newcomers to this mailing list - this has been discussed multiple > times in the past 10+ years) > > ACS 4.17 has introduced routed mode for IPv6 in Isolated networks and VPCs, > we are currently working on extending that to IPv4 as well, which will > support the current NAT'ed mode and also a routed mode (inspired by the NSX > integration > https://urldefense.com/v3/__https://www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$<https://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$><https://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e><https://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3e><https://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3e%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3chttps://urldefense.com/v3/__https:/www.youtube.com/watch?v=f7ao-vv7Ahk__;!!P9cq_d3Gyw!gRe7Js-1plXE8vRRc_mJQIri5T4-Z1zOFVmqEwmHGE_AGkN6P6BU5T8nq0WL4Fx0MTwP0p-ucEL6DjwFzB7TaoBNnS4$%3e%3e%3e> > ). > > With stock ACS (i.e. without NSX or OpenSDN) this routing is purely static, > with the operator being responsible to add static routes to the Isolated > network or VPC tiers via the "public" (outside) IP of the virtual router. > > The next step on this journey is to add some kind of dynamic routing. One way > that I have in mind is using dynamic BGP: > > 1 - Operator configures one or more BGP peers for a given Zone (with > different metrics) > 2 - Operator presents a pool of Private AS numbers to the Zone (just > like we do for VLANs) > 3 - When a network is created with an offering which has dynamic > routing enabled an AS number is allocated > 4 - ACS configures the BGP session on the VR, advertising all its > connected networks > > This way there's no need to reconfigure the upstream router for each > new ACS network (it just needs to allow dynamic BGP peering from the > pool of AS numbers presented to the zone) > > This implementation could also be used for Shared Networks, in which case the > destination advertised via BGP is to the gateway of the shared network. > > There could also be an offering where we allow for end users to setup the BGP > parameters for their Isolated or VPC networks, which can then peer with > upstream VNF(s). > > Any and all input is very welcome... > > Taking the liberty to tag some of you: @Wei > Zhou<mailto:wei.z...@shapeblue.com> @Wido den > Hollander<mailto:w...@widodh.nl> @Kristaps > Čudars<mailto:kristaps.cud...@telia.lv> > > Cheers, > Alex