[rbridge] draft-ietf-trill-prob-00.txt
Donald.Eastlake at motorola.com (Eastlake III Donald-LDE008) Fri, 18 August 2006 18:33 UTC
From: "Donald.Eastlake at motorola.com"
Date: Fri, 18 Aug 2006 14:33:02 -0400
Subject: [rbridge] draft-ietf-trill-prob-00.txt
Message-ID: <3870C46029D1F945B1472F170D2D9790014828B7@de01exm64.ds.mot.com>
Hi, Here are a few comments from me as a WG member on the draft-ietf-trill-prob-00.txt draft: Page 2 in abstract at the top: I don't actually think of typical distance vector routing as much, if any, better that spanning tree, both being substantially inferior to link state protocols. I would suggest, in the 3rd & 4th line, changing "apply network layer routing (e.g., link state or distance vector)" to "apply modern network layer routing (e.g., link state)". Section 2.2, page 6, first question: Sure, just make a ring. STP will turn off some arbitrary link in it (actually determined by MAC addresses or something). Now imagine that a link approximately opposite the one turned off by STP fails. The direction towards root will change for all or almost all nodes, those that were few hops apart will be many hops apart and vice versa... Section 2.2, page 6, second question: I'm not an expert but I have heard things said that imply that the worst case RSTP settling time is related to 3 times the network diameter, where diameter is the maximum of the minimum number of hops between all pairs of nodes. Section 2.2, page 6, third question: I'm not sure if this is an answer to your question but it seems to me that bridge port learning has problems solvable by link state protocol distribution of the information. A---\ B---\ C1--X--/ \ D---...Z C2--Y--\ / E---/ F---/ Assuming node C moves from C1 to C2. There had been traffic from C1 to Z. Traffic from C2 to Z would presumably cause D, E, and Y to learn of its new attachment, changing information at D. But is there any mechanism in spanning tree systems whereby B's opinion of how to forward frames to C would be updated? What happened if there is a spurt of data from A addressed to C? If B/D/E were Rbridges, presumably information about C's new attachment would have gotten to B. Section 2.3: given that this is about "Robustness to Link Interruption", I suppose it is reasonable that it talks so much about alternative paths. But I think Section 2 is too heavy on alternative paths and very odd that it does not mention loops. Perhaps 2.3 should be "Robustness to Link Interruption and Appearance" and could mention loops due to a repeater coming up or the like. Section 2.4, page 7, last paragraph: I don't think multipath is "the focus of TRILL". Maybe optimized paths is. Also, should this paragraph mention loops? Section 2.5: I found this section, or at least the first part of it, a bit hard to understand. Are the first two bullet items actually related to "deployment", mentioned before them, or to "larger scale", which is not mentioned until after them? Section 3.2, page 9: Suggest dropping "(inadvertent)". Section 3.3: Spanning tree tries to avoid loops but the appearance of a link can create one. Section 3.4: I'm not sure that the "TRILL solution" is necessarily more "stable". It is supposed to provide more optimized paths and be more robust... Section 3.5: I believe this has to do with things like a cloud of TRILL devices, bridge, and hubs having "multiple attachments" of various sorts such as to the Internet and questions about which is chosen for various traffic, how you can be sure that a broadcast coming in through one attachment isn't sent back out on another so it doesn't loop, etc. I believe this sort of concern is why the TRILL charter includes the following: "The TRILL working will work with the L2VPN WG to develop interworking between TRILL and 802.1D bridges at the edge, such that a bridged sub-cloud could be attached to TRILL devices in more than one place for redundancy." Section 5, page 13, 4th paragraph: Suggest inserting "or adding" after "interrupting". References: "D" should be capitalized in 802.1D. Document title is "IEEE Standard for Local and metropolitan area networks: Media Access Control (MAC) Bridges" dated 9 June 2004. 802.1Q, 7 May 2003, is "IEEE Standards for Local and metropolitan area networks: Virtual Bridged Local Area Networks" and incorporates 802.1v and 802.1s. In 802-land, the capitalization of the alphabetic suffix makes a difference. If it is upper case, you are talking about a free-standing document. If it is lower case, then you are talking about an amendment to a base document which will get rolled into that base document the next time it is updated. Thanks, Donald
- [rbridge] draft-ietf-trill-prob-00.txt Eastlake III Donald-LDE008
- [rbridge] draft-ietf-trill-prob-00.txt Russ White
- [rbridge] draft-ietf-trill-prob-00.txt Eastlake III Donald-LDE008