[rbridge] Additioanla clarification Abridges port types Re: A question about (R)STP and routing protocol / requestfor comments to Abridges draft

[gibanez at it.uc3m.es ( Guillermo Ibáñez )]

Hi Suping,
   Just a clarification. Abridges have core ports (those connected to 
other Abridges) and distribution ports (those connected to standard 
bridges). Core ports run AMSTP protocol and distribution ports run RSTP 
or STP protocol. The ports selfconfigure as core or distribution ports 
according to the standard protocol migration procedure 802.1D (i.e. 
depending on the protocol type of STP BPDUs they hear from neighbour).
Distribution ports are standard ports regarding learning. Specific 
operation corresponds to core ports.
Regards
Guillermo
 
Suping Zhai wrote:

>Hi Guillermo,
>Thank you for your response. After I have gone through the following draft, there are some doubts wrt the following aspects,
>1) what's the orientation wrt the basic RSTP instance in the core Abrige network?  What I understand is per-Abridge RSTP is enough for the forwarding and there is no need for the aforementioned basic RSTP instance.
>
>2) In 4.3.2 it says that Abridge may learn from the received frames both outer MAC and inter MAC address, called the double MAC adress learning. On the other hand, when talk about the symmetric spanning tree problem in section 4.8 and 6.2, it says that Abridges don't learn addresses. So which is the right one or I missed some points?
>
>Regards
>Suping  
>  
>
>>Let me give my opinions interleaved.
>>By the way, as the Abridges draft we submitted last december is 
>>somewhere between the current RBridges proposal and the IEEE Shortest 
>>Path Bridging, we think it would help clarify to obtain comments to our 
>>draft from the RBridges WG members.
>>Regards
>>Guillermo Ibanez
>>
>>
>>http://www.ietf.org/internet-drafts/draft-gibanez-trill-abridge-00.txt
>>
>>Guillermo Ibanez
>>
>>
>>Radia Perlman wrote:
>>
>>    
>>
>>>Suping Zhai,
>>>
>>>You are not the first person to ask these questions.
>>>There is a lot of confusion over what is different between RSTP and STP,
>>>and what is different between RSTP and link state routing.
>>>I'll try to explain.
>>>
>>>1) What's the difference between RSTP and STP?
>>>
>>>RSTP and STP are almost identical, are compatible,
>>>and it would be less confusing to think of RSTP as a different
>>>version of the bridge spec than a different protocol.
>>>
>>>RSTP calculates the
>>>same spanning tree, and uses the same algorithm and even
>>>packet formats as STP. Both RSTP and STP calculate a tree of shortest paths
>>>      
>>>
>>>from the Root bridge, which is the bridge with lowest ID/priority.
>>    
>>
>>>The major difference between RSTP and STP
>>>is how they avoid temporary loops.
>>>STP did it with a timer. RSTP coordinates between neighbors
>>>to turn on links more quickly after topology changes.
>>>
>>>However, in either case, it is impossible to always avoid temporary
>>>loops...the simplest case is when a repeater comes up, or if too many
>>>spanning tree messages get lost due to congestion.
>>>
>>>So to summarize, both STP and RSTP use the same basic
>>>algorithm...the heart of algorithm is to calculate a tree
>>>of shortest paths from a single point.
>>>And the resulting tree and data packet
>>>forwarding path is the same in both (because it is the
>>>same algorithm). A single shared loop-free subset
>>>of the physical topology is calculated, upon which data packets are
>>>forwarded.
>>>
>>> 
>>>
>>>2) What is the difference between RSTP and IS-IS?
>>>
>>>This is harder to answer than the difference between RSTP and STP,
>>>because IS-IS and spanning tree (RSTP/STP) are so very different from
>>>each other. IS-IS passes around topology
>>>information so that every RBridge calculates the shortest path between
>>>itself
>>>and each
>>>destination. With spanning tree (RSTP/STP) each bridge just knows which
>>>subset of its
>>>own ports are "in" or "out" of the spanning tree. If a link is not in
>>>the spanning tree,
>>>then it cannot be used, even if it is the shortest path between point A
>>>and B, and
>>>pairwise paths can be quite suboptimal. Furthermore, traffic is concentrated
>>>on the links in the spanning tree, because no traffic can be on links
>>>not in the
>>>spanning tree.
>>>
>>>Also spanning tree bridges learn, based on
>>>seeing data traffic, which
>>>direction the source is. (which of its own local ports lead to the
>>>source of
>>>the data packet). So packets must always arrive from a particular source
>>>from
>>>the same direction or else bridges will get confused. In contrast, with
>>>IS-IS,
>>>switches can do path splitting; using multiple paths to reach a destination.
>>>
>>> 
>>>
>>>      
>>>
>>It is important to remember that the standard convergence times of RSTP 
>>are lower than those of routing protocols, unless specific measures for 
>>milliseconds convergence are applied.
>>
>>    
>>
>>>3) What is difference between RSTP and RBridge approach?
>>>
>>>RBridge incorporates
>>>a TTL into the header of forwarded data packets so a temporary loop is
>>>not really bad, so it need not
>>>be very conservative about avoiding loops. Also,
>>>a link state protocol (IS-IS) calculates shortest
>>>pair-wise paths. Also, it can calculate multiple paths to the
>>>destination, and do path splitting.
>>> 
>>>
>>>      
>>>
>>Both IEEE Shortest Path Bridging and the Abridges draft proposal ( see
>>http://www.ietf.org/internet-drafts/draft-gibanez-trill-abridge-00.txt)
>>obtain shortest paths. They do not allow  path splitting, but are less 
>>complex than IS-IS
>>and do not mix STP and IS-IS protocols in same area.
>>
>>    
>>
>>>4) What is the difference between recent work in IEEE to calculate
>>>per-ingress trees, and RBridge approach?
>>>
>>>IEEE may indeed wind up doing the same thing as RBridge. Several years
>>>before TRILL I tried to get IEEE interested in doing this approach,
>>>but they were not interested, perhaps because I didn't explain it
>>>well enough. Now that work has started in IETF, I think it is possible
>>>that IEEE will converge on the same solution, which would actually
>>>be good for the industry.
>>>Radia
>>> 
>>>
>>>      
>>>
>>I do not see an easy convergence between IEEE and RBridge proposals 
>>unless the two groups cooperate strongly.
>>IEEE will likely preserve basic RSTP mechanisms (fast convergence and 
>>distance vector to (multiple) root bridges) with multiple spanning trees 
>>(this is my guess, I do not follow their recent activities), while 
>>RBridge uses IS-IS (link state mechanisms) as basic protocol.
>>
>>
>>Guillermo Ibanez
>>
>>    
>>
>>>Suping Zhai wrote:
>>>
>>> 
>>>
>>>      
>>>
>>>>hi,
>>>>I am confronted with some questions when reading the Rbridge protocol related draft and hope someone can help me with details.
>>>>What's the essential difference between (R)STP and routing protocol(e.g. IS-IS)? 
>>>>How dose the routing protocol(e.g. IS-IS) can optimize the routing path while (R)STP can't? 
>>>>Can we optimize the (R)STP protocol itself to reach that goal? If that's the case, then I think that all RBridge WG works can be substituted. And I have seen some work been done in IEEE802.1aq which is on the way to the per bridge MSTP road. But what I imagine is that should we get a comparable simple and untangled solution for the bridge network routing path?
>>>>
>>>>TIA.
>>>>
>>>>Best Regards,
>>>>zhaisuping at huawei.com
>>>>Huawei Technologies Co., Ltd.
>>>>Tel: [phone]
>>>>Fax: [phone]
>>>>2006-02-23
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>------------------------------------------------------------------------
>>>>
>>>>_______________________________________________
>>>>rbridge mailing list
>>>>rbridge at postel.org
>>>>http://www.postel.org/mailman/listinfo/rbridge
>>>>
>>>>
>>>>   
>>>>
>>>>        
>>>>
>>> 
>>>
>>>------------------------------------------------------------------------
>>>
>>>_______________________________________________
>>>rbridge mailing list
>>>rbridge at postel.org
>>>http://www.postel.org/mailman/listinfo/rbridge
>>> 
>>>
>>>      
>>>
>>-- 
>>Guillermo Ib??ez
>>Departamento de Ingenier?a Telem?tica
>>Universidad Carlos III de Madrid
>>1.1.B.11 Colmenarejo 91-6241393
>>4.1.F.13 Legan?s 91-6248794
>>
>>_______________________________________________
>>rbridge mailing list
>>rbridge at postel.org
>>http://www.postel.org/mailman/listinfo/rbridge
>>    
>>
>
>
>
>  
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gibanez.vcf
Type: text/x-vcard
Size: 424 bytes
Desc: not available
Url : http://www.postel.org/pipermail/rbridge/attachments/20060228/21c4a45e/gibanez.vcf