[rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
Jon Hudson
jon.hudson at gmail.com
Sat Jan 14 20:10:19 PST 2012
Just to add a few data points
As of now only Cisco and Brocade have been shipping a "TRILL based" product
By late 2011 numbers of units in production networks
Cisco >500
Brocade >300
So with growth let's assume 1000-1200 total.
However neither Brocade or Ciscos solution is pure openTRILL. Both have work to do on that front.
As I understand it Force10 & Extreme will ship soon but have not yet.
In addition the affinity-TLV does not require a hardware change for anyone (that I am aware of). And since those ~1000 today will need some software updates anyway to become compliant, backwards comparability is essentially a non issue for the affinity TLV.
If the PSN-LSP solution increased scalability, or added functionality then it (or anything else that does) deserves careful consideration.
PSN-LSP seems to only gain something that isn't hyper critical in this case and adds scaling concerns.
I often have to alleviate concerns that TRILL contains scaling issues in its use of IS-IS, I don't wish to add fuel to that fire.
My 0x2 bits
On Jan 14, 2012, at 7:04 AM, "Tissa Senevirathne (tsenevir)" <tsenevir at cisco.com> wrote:
> Hi Hongjun
>
> Technically speaking what you are proposing is similar to Affinity TLV approach. With the difference of using Pseudonode LSP instead of the Affinity TLV .
>
> Main issue with PSN-LSP approach is the scale, Let me explain that in bit more details.
>
> Affinity TLV is a single sub-TLV announced by each of the members of active-active pair or pairs.
>
> Similarly PSN-LSP is also announced by each of the members of active-active pair or pairs.
>
> However, PSN-LSP is a fully contained LSP that contain multiple TLV s that needed to fully form a PSN.
>
> This will increase the size of the LSP-DB of the campus and introduce X more additional TLVs.
>
> There can be many active-active pair/pairs in the network, and with this approach we have so many more additional LSP in the network.
>
> As we know size of LSP-DB is a major scaling issue and always want to keep it as small as possible, where possible.
>
> Lets also look at why PSN-LSP approach being proposed. It is to solve a possible backward compatibility issue.
>
> TRILL still in infancy and there are not many TRILL networks. So the backward compatibility is not a wide spread issue. Also affinity draft has backward compatibility mode.
>
> What we should look at is having a solution that help us to scale better as TRILL become widely deployed in Large data-centers/Enterprises where such active-active edge is widely used.
>
> With PSN-LSP approach I think we are unnecessarily injecting additional LSP and TLVs and making it more complicated and less scalable when we can solve with a single sub-TLV. Also Affinity TLV draft discuss backward compatibility approach, in the event there older Rbridges.
>
> Thanks
> Tissa
>
> From: zhai.hongjun at zte.com.cn [mailto:zhai.hongjun at zte.com.cn]
> Sent: Friday, January 13, 2012 8:17 PM
> To: Janardhanan Pathangi Narasimhan
> Cc: rbridge at postel.org; rbridge-bounces at postel.org; Tissa Senevirathne (tsenevir)
> Subject: Re: [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
> Hi Jana, Jon and Tissa,
>
> > I think your proposal is that form a tree rooted at RB1 and another tree rooted at RB2. The tree rooted at RB1 would have RBV-RB1 link,
> > and similarly the tree rooted at RB2 would have RB2-RBV link. Hence you are forcing the affinity of RB1 or RB2 to RBV by placing the root on the RBridges themselves.
> > Is that correct?
>
> No, It didn't mean that the trees must rooted at RB1, RB2, respectively. It has no any needs of where the roots of these trees are. These trees are ordinary trees calculated by all RBridges in TRILL campus wide.
>
> For example, there are 4 actually-calculated trees in TRILL campus, T1 through T4. T1 roots at RBi, T2 roots at RBj, T3 roots at RBm and T4 roots at RBn. If T1 and T2 are assigned to member RBridge RB1, and the rest trees to RB2. The method used to assign trees to a member RBridge is the same as given in Section 5.1 in draft CMT.
>
> In the below topology, Since RB1 knows it is a member RBridge of Rbridge Group RBv1 and that T1/T2 are assigned to it, RB1 can originate a pseudonode LSP(even if it has not seen any neighbors in RBv by hellos) and advertize it in TRILL campus. This pseudonode LSP contains a Nickname sub-TLV where nickname of RBv1 is carried, and a Trees sub_TLV where the wanted-to-using tress(i.e., T1 and T2) are carried, as well as a neighbor TLV where RB1 is the only neighbor contained in this pseudonode LSP.
>
> After receiving this psuedonode LSP, another RBridge, e.g., RBx, knows the paths to RBv through RB1 on T1 and T2. Then RPF information for RBv1 can be calculated correctly by RBx on these two trees.
> RBx
> |
> Trill Campus
> | |
> +----------+ +------------+
> | RB1 | | RB2 |
> +----------+ + ------------+
> \ /
> \ /
> \ /
> ---------- -----------
> > | | < Port channel (LAG)
> | |
> +--------------------+
> | CE |
> +---------------------+
>
> Besides RBv1, If RB1joins another Rbridge Group RBv2, and T2/T3 are assigned to RB1, RB1 can originates another pseudonode LSP, where nickname of RBv2 and T2/T3 are carried in Nickname sub-TLV and Trees sub-TLV, respectively. After receiving this LSP, RBx can knows path to RBv2 through RB1 on T2 and T3. RPF check information can also calculated correctly by RBx on thes two trees.
>
> Similar things RB2 can do, and RPF check information can be calculated correctly by RBx on the assigned trees
>
> > Yes, this mechanism would work, but would not scale up. For e.g. in the network, if we have 10 such pairs of RB, i.e. RB1-RB2 providing RBV1, RB3-RB4 providing RBV2,
> > and so on, then to ensure that RB1-RBV1, RB2-RBV1, RB3-RBV2, RB4-RBV2 and so on are present, we will have to minimum build 20 trees (10 pairs of RB).
> > This will not scale, while with affinity TLV irrespective of number of such RB pairs which are providing such services we would need only two trees.
>
> No, It does not require much more trees to build if using pseudonode LSP, instead of affinity TLV, to notify other RBridges the affinity of RBv via RB1. It just needs to originate much more pseudonode LSPs if an Rbridge joins more than one RBridge Group.
>
>
>
> Thanks,
> Zhai Hongjun
> """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
> Protocol Development Dept.VI, Central R&D Institute, ZTE Corporation
> No. 68, Zijinghua Road, Yuhuatai District, Nanjing, P.R.China, 210012
>
> Zhai Hongjun
>
> Tel: +86-25-52877345
> Email: zhai.hongjun at zte.com.cn
> """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
>
>
>
>
> Janardhanan Pathangi Narasimhan <jana at force10networks.com>
> 发件人: rbridge-bounces at postel.org
> 2012-01-13 21:48
>
> 收件人
> "zhai.hongjun at zte.com.cn" <zhai.hongjun at zte.com.cn>, "Tissa Senevirathne (tsenevir)" <tsenevir at cisco.com>
> 抄送
> "rbridge at postel.org" <rbridge at postel.org>, "rbridge-bounces at postel.org" <rbridge-bounces at postel.org>
> 主题
> Re: [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
>
>
> Hi Hongjun,
>
> I think your proposal is that form a tree rooted at RB1 and another tree rooted at RB2. The tree rooted at RB1 would have RBV-RB1 link, and similarly the tree rooted at RB2 would have RB2-RBV link. Hence you are forcing the affinity of RB1 or RB2 to RBV by placing the root on the RBridges themselves. Is that correct?
>
> Yes, this mechanism would work, but would not scale up. For e.g. in the network, if we have 10 such pairs of RB, i.e. RB1-RB2 providing RBV1, RB3-RB4 providing RBV2, and so on, then to ensure that RB1-RBV1, RB2-RBV1, RB3-RBV2, RB4-RBV2 and so on are present, we will have to minimum build 20 trees (10 pairs of RB). This will not scale, while with affinity TLV irrespective of number of such RB pairs which are providing such services we would need only two trees.
>
> Thanks
> Jana
>
>
> From: rbridge-bounces at postel.org [mailto:rbridge-bounces at postel.org] On Behalf Of zhai.hongjun at zte.com.cn
> Sent: Friday, January 13, 2012 10:35 AM
> To: Tissa Senevirathne (tsenevir)
> Cc: rbridge at postel.org; rbridge-bounces at postel.org
> Subject: Re: [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
> Hi Tissa，
>
> In the diagram you given, assume 4 distribution trees are acturally calculated in TRILL campus, i.e., T1 through T4. T1 and T2 are assigned to RB1, T3 and T4 are assigned to RB2.
>
> Even if RB1 can not see RB2 by hellos, (I think that) RB1 can generate a pseudonode LSP, Pseu-LSP-RB1, to announce the virtual RBridge (RBv) nickname and the trees assigned to it, i.e, T1 and T2. In this pseudonode LSP, RB1 is the only one neighbor (or RB1 and the virtual RBridge RBv are the only two neigbors) of this pseudonode. After receiving this pseudonode LSP, RBx can know the path to pseudonode RBv through RB1 on T1 or T2, shown in the following diagram.
>
> RBx
> |
> Trill Campus
> | |
> +----------+ +------------+
> | RB1 | | RB2 |
> +----------+ + ------------+
> \
> / \
> /RBv \
> ---------
> scheme of T1 or T2
>
> RB1 uses Nickname Sub-TLV and Trees Sub-TLV (instead of the Affinity sub-TLV) in Pseu-LSP-RB1, to contains the RBv's nickname and the assigned trees T1 and T2.
>
> RB2 also annouce similar informtion in its pseudonode LSP, then RBx can know the path to RBv through RB2 on T3 and T4.
>
> RBx
> |
> Trill Campus
> | |
> +----------+ +------------+
> | RB1 | | RB2 |
> +----------+ + ------------+
> \ /
> \ /
> \ /
> ---------- -----------
> > | | < Port channel (LAG)
> | |
> +--------------------+
> | CE |
> +---------------------+
>
>
>
>
> Thanks,
> Zhai Hongjun
> """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
> Protocol Development Dept.VI, Central R&D Institute, ZTE Corporation
> No. 68, Zijinghua Road, Yuhuatai District, Nanjing, P.R.China, 210012
>
> Zhai Hongjun
>
> Tel: +86-25-52877345
> Email: zhai.hongjun at zte.com.cn
> """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
>
>
> "Tissa Senevirathne (tsenevir)" <tsenevir at cisco.com>
> 2012-01-13 12:21
>
>
> 收件人
> <zhai.hongjun at zte.com.cn>
> 抄送
> <rbridge at postel.org>, <rbridge-bounces at postel.org>
> 主题
> RE: [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
>
>
>
>
>
>
> Hi Hongjun
>
> I am not quite clear with the way you are proposing to use assigned nicknames in the pseudonode LSP.
>
> Can you please us the diagram below and explain how your proposal will be used.
>
> CE means in the following diagram is an 802.1D device that is attached to RB1 and RB2 through Pt-Pt links and treated as a port channel from CE stand point.
>
> RBx
> |
> Trill Campus
> | |
> +----------+ +------------+
> | RB1 | | RB2 |
> +----------+ + ------------+
> \ /
> \ /
> \ /
> ---------- -----------
> > | | < Port channel (LAG)
> | |
> +--------------------+
> | CE |
> +---------------------+
>
>
> From: zhai.hongjun at zte.com.cn [mailto:zhai.hongjun at zte.com.cn]
> Sent: Thursday, January 12, 2012 7:59 PM
> To: Tissa Senevirathne (tsenevir)
> Cc: rbridge at postel.org; rbridge-bounces at postel.org
> Subject: RE: [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
> Hi Tissa,
>
> I have read your draft-tissa-trill-cmt-00.txt with great interests. In this draft, a new TLV, Affinity Sub-TLV, is introduced to give a member RBdige of a virtual RBridge group much flexibity in describing the virtual RBridge nickname and specifying the path of association to specific distribution tree(S). However, this new TLV causes some backward compatibilty problems between pre Affinity sub-TLV RBridges and new RBridges. Even if there is only one pre Affinity sub-TLV RBridge in TRILL network,the RPF check for virtual Rbridge will fail on this RBridge.
>
> Is the Affinity Sub-TLV necessary for a member RBridge to announce the virtual Rbridge nickname and the specific distribution trees? Maybe it is not.
>
> In a given vritual group, each member Rbridge can advertise a specific pseudonode-LSP to annouce the virtual RBridge nickname and the trees assigned to it.The nickname and the assigned trees are contained in the given Nickname Sub-TLV and Trees Sub-TLV in RFC6326. This pseudonode-LSP represents a special virtual link, on which there is only one RBridge, i.e., the RBridge advertising this LSP. Only if different member Rbridge is assigned different tree(s), RPF check information for the virtual Rbrigde in different trees can be correctly calculated on all the Rbridges in TRILL network, then the member RBridge can safely use the virtual Rbridge nickname to do the necessary TRILL-encapsultion on received native frames.
>
> Therefore, backward compatibilty problems, caused by Affinity sub-TLV, is removed, while the similar flexibilty of Affinity sub-TLV remains.
>
>
> Thanks,
> Zhai Hongjun
> """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
> Protocol Development Dept.VI, Central R&D Institute, ZTE Corporation
> No. 68, Zijinghua Road, Yuhuatai District, Nanjing, P.R.China, 210012
>
> Zhai Hongjun
>
> Tel: +86-25-52877345
> Email: zhai.hongjun at zte.com.cn
> """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
>
> "Tissa Senevirathne (tsenevir)" <tsenevir at cisco.com>
> 发件人: rbridge-bounces at postel.org
> 2012-01-06 10:11
>
>
>
>
> 收件人
> <rbridge at postel.org>
> 抄送
> 主题
> [rbridge] Coordinated Multicast Trees to solve RPF issues on active-active edge and more
>
>
>
>
>
>
>
>
>
>
>
>
> Dear All
>
> We have published draft-tissa-trill-cmt-00.txt, in this document we present a flexible framework to associate child RBridges to different parent RBRidges for RPF purpose.
>
> Active-active forwarding at the TRILL edge is a very important feature. Representation of the active-active sub system with a single virtual RBRidge is a common practice to achieve load balancing and to avoid MAC flip flop. However, use of virtual RBridge lead to RPF failures. In this document we provide a solution to address the RPF failures. Please review and share comments.
>
> ID can be found in URL: http://www.ietf.org/id/draft-tissa-trill-cmt-00.txt
>
> Summary:
>
> Filename: draft-tissa-trill-cmt
> Revision: 00
> Title: Coordinated Multicast Trees (CMT)for TRILL
> Creation date: 2012-01-05
> WG ID: Individual Submission
> Number of pages: 14
>
> Abstract:
> TRILL facilitates loop free connectivity to non TRILL legacy
> networks via choice of an Appointed Forwarder for set of VLANs.
> Appointed Forwarder provides VLAN level load sharing with active-
> standby model. Mission critical operations such as High Performance
> Data Centers require active-active load sharing model. Active-Active
> load sharing model can be accomplished by representing any given non
> TRILL legacy network with a single virtual RBridge. Virtual
> representation of the non-TRILL legacy network with a single RBridge
> poses serious challenges in multi-destination RPF calculations. This
> document presents the required enhancements to build Coordinated
> Multicast Trees (CMT) within the TRILL campus. CMT provides
> flexibility to RBridges to select desired path of association to a
> given distribution tree.
>
>
> Thanks
> Tissa_______________________________________________
> rbridge mailing list
> rbridge at postel.org
> http://mailman.postel.org/mailman/listinfo/rbridge
>
>
> --------------------------------------------------------
> ZTE Information Security Notice: The information contained in this mail is solely property of the sender's organization. This mail communication is confidential. Recipients named above are obligated to maintain secrecy and are not permitted to disclose the contents of this communication to others.
> This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the originator of the message. Any views expressed in this message are those of the individual sender.
> This message has been scanned for viruses and Spam by ZTE Anti-Spam system.
>
> --------------------------------------------------------
> ZTE Information Security Notice: The information contained in this mail is solely property of the sender's organization. This mail communication is confidential. Recipients named above are obligated to maintain secrecy and are not permitted to disclose the contents of this communication to others.
> This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the originator of the message. Any views expressed in this message are those of the individual sender.
> This message has been scanned for viruses and Spam by ZTE Anti-Spam system._______________________________________________
> rbridge mailing list
> rbridge at postel.org
> http://mailman.postel.org/mailman/listinfo/rbridge
>
>
>
> --------------------------------------------------------
> ZTE Information Security Notice: The information contained in this mail is solely property of the sender's organization. This mail communication is confidential. Recipients named above are obligated to maintain secrecy and are not permitted to disclose the contents of this communication to others.
> This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the originator of the message. Any views expressed in this message are those of the individual sender.
> This message has been scanned for viruses and Spam by ZTE Anti-Spam system.
> _______________________________________________
> rbridge mailing list
> rbridge at postel.org
> http://mailman.postel.org/mailman/listinfo/rbridge
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.postel.org/pipermail/rbridge/attachments/20120114/cf190943/attachment-0001.html
More information about the rbridge
mailing list