Thread: [mpls-linux-devel] VPLS support?
Status: Beta
Brought to you by:
jleu
Menu
▾
▴
mpls-linux-devel
|
From: David F. <da...@ab...> - 2010-04-22 16:00:20
|
Hi, What is the situation with VPLS support? In particular, I believe that mpls-linux's existing Ethernet over MPLS works by virtue of hooking on ebtables. That is a case of modifying one ethernet packet on the way through. Whereas, in a scenario where there are three LERs (forming a VPLS), if an ethernet packet arrived from outside the VPN, then the LER would need to send two packets - one to each of the other two LERs in the VPLS VPN - each with a different MPLS label. Is there any kernel support for this sort of distribution in mpls-linux/ebtables/elsewhere? I see from the archives that Pali was doing some work on this area late last year, although discussion seemed to revolve mainly around the signalling plane (quagga etc) and not this dataplane issue. Thanks in advance, David Fraser |
|
From: <jl...@mi...> - 2010-04-22 16:25:48
|
David, The multi-point behavior you discuss below has not been implemented. My thinking at this time is that we can use the linux bridging core to accomplish this. For this to occur a VPLS LSP needs to be represented to the bridge core as an ethernet interface. This could be accomplished with a modified version of the mpls tunnel driver. Due to the recent activity, I've picked up the project again. I'm/we're in the process of getting all the fuctionality working on a 2.6.32 kernel. Once I get that accomplished I think this would be a good place to do some additional work. On Thu, Apr 22, 2010 at 04:54:47PM +0100, David Fraser wrote: > > Hi, > > What is the situation with VPLS support? > > In particular, I believe that mpls-linux's existing Ethernet over MPLS > works by virtue of hooking on ebtables. That is a case of modifying one > ethernet packet on the way through. Whereas, in a scenario where there > are three LERs (forming a VPLS), if an ethernet packet arrived from > outside the VPN, then the LER would need to send two packets - one to > each of the other two LERs in the VPLS VPN - each with a different MPLS > label. Is there any kernel support for this sort of distribution in > mpls-linux/ebtables/elsewhere? > > I see from the archives that Pali was doing some work on this area late > last year, although discussion seemed to revolve mainly around the > signalling plane (quagga etc) and not this dataplane issue. > > Thanks in advance, > > David Fraser > > ------------------------------------------------------------------------------ > _______________________________________________ > mpls-linux-devel mailing list > mpl...@li... > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel -- James R. Leu jl...@mi... |
|
From: Rodrigo D. <ro...@tr...> - 2010-04-22 18:31:38
|
When I made some tests with ebtables, I got some kernel panics, but I not investigated why because I mainly interested on L3VPNs. Renato's patch to 2.6.32 doesn't contains the ebtables support too. You need to use 2.6.27 from git. On Thu, Apr 22, 2010 at 1:25 PM, <jl...@mi...> wrote: > David, > > The multi-point behavior you discuss below has not been implemented. > My thinking at this time is that we can use the linux bridging > core to accomplish this. For this to occur a VPLS LSP needs to > be represented to the bridge core as an ethernet interface. > This could be accomplished with a modified version of the > mpls tunnel driver. > > Due to the recent activity, I've picked up the project again. > I'm/we're in the process of getting all the fuctionality working on > a 2.6.32 kernel. Once I get that accomplished I think this > would be a good place to do some additional work. > > On Thu, Apr 22, 2010 at 04:54:47PM +0100, David Fraser wrote: > > > > Hi, > > > > What is the situation with VPLS support? > > > > In particular, I believe that mpls-linux's existing Ethernet over MPLS > > works by virtue of hooking on ebtables. That is a case of modifying one > > ethernet packet on the way through. Whereas, in a scenario where there > > are three LERs (forming a VPLS), if an ethernet packet arrived from > > outside the VPN, then the LER would need to send two packets - one to > > each of the other two LERs in the VPLS VPN - each with a different MPLS > > label. Is there any kernel support for this sort of distribution in > > mpls-linux/ebtables/elsewhere? > > > > I see from the archives that Pali was doing some work on this area late > > last year, although discussion seemed to revolve mainly around the > > signalling plane (quagga etc) and not this dataplane issue. > > > > Thanks in advance, > > > > David Fraser > > > > > ------------------------------------------------------------------------------ > > _______________________________________________ > > mpls-linux-devel mailing list > > mpl...@li... > > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel > > -- > James R. Leu > jl...@mi... > > > ------------------------------------------------------------------------------ > > _______________________________________________ > mpls-linux-devel mailing list > mpl...@li... > https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel > > |
|
From: David F. <da...@ab...> - 2010-04-23 14:58:20
|
James, Thanks. Very interesting. Glad you are picking up the baton again. Just to complicate things further :-)... not that I know much about network namespaces... but would this sort of approach also be compatible with network namespaces, so that you could simulate tunnelling across a VPLS network on just the one physical Linux machine!? I had a bit of a google about namespaces, ebtables and bridging and couldn't find any evidence that bridging was within the namespace umbrella, therefore it wasn't clear that any label switching hooking off this would be appropriate for the given "virtual brouter". Any ideas? (I know that VRF is up your street). Hope that makes sense. Thanks again, David On 22/04/2010 17:25, jl...@mi... wrote: > David, > > The multi-point behavior you discuss below has not been implemented. > My thinking at this time is that we can use the linux bridging > core to accomplish this. For this to occur a VPLS LSP needs to > be represented to the bridge core as an ethernet interface. > This could be accomplished with a modified version of the > mpls tunnel driver. > > Due to the recent activity, I've picked up the project again. > I'm/we're in the process of getting all the fuctionality working on > a 2.6.32 kernel. Once I get that accomplished I think this > would be a good place to do some additional work. > > On Thu, Apr 22, 2010 at 04:54:47PM +0100, David Fraser wrote: >> >> Hi, >> >> What is the situation with VPLS support? >> >> In particular, I believe that mpls-linux's existing Ethernet over MPLS >> works by virtue of hooking on ebtables. That is a case of modifying one >> ethernet packet on the way through. Whereas, in a scenario where there >> are three LERs (forming a VPLS), if an ethernet packet arrived from >> outside the VPN, then the LER would need to send two packets - one to >> each of the other two LERs in the VPLS VPN - each with a different MPLS >> label. Is there any kernel support for this sort of distribution in >> mpls-linux/ebtables/elsewhere? >> >> I see from the archives that Pali was doing some work on this area late >> last year, although discussion seemed to revolve mainly around the >> signalling plane (quagga etc) and not this dataplane issue. >> >> Thanks in advance, >> >> David Fraser >> >> ------------------------------------------------------------------------------ >> _______________________________________________ >> mpls-linux-devel mailing list >> mpl...@li... >> https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel > -- David Fraser Abelon Systems Ltd 0131 661 9088 <-- 01506 638 915 |
×
Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.
Thanks for helping keep SourceForge clean.
X