Re: Working group last call: draft-ietf-ccamp-pc-spc-rsvpte-ext-02
Lou Berger <lberger@labn.net> Fri, 01 May 2009 17:24 UTC
Envelope-to: ccamp-data0@psg.com
Delivery-date: Fri, 01 May 2009 17:26:15 +0000
DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=default; d=labn.net; h=Received:Message-ID:Date:From:User-Agent:MIME-Version:To:CC:Subject:References:In-Reply-To:Content-Type:Content-Transfer-Encoding:X-Identified-User; b=mVpwOnfv6GLSmoqIFAhjBpYwM4namIdUsR+wl61lAJYAJmazJYMGiuG5ZZ1Rv9L1rXuGwBsoiQGn3IDU64MQUm1CHCKS1/FrwWlt2eDlr1htnmIT09JSJAV0i1yopvVy;
Message-ID: <49FB303C.4060408@labn.net>
Date: Fri, 01 May 2009 13:24:12 -0400
From: Lou Berger <lberger@labn.net>
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.1b3pre) Gecko/20090408 Eudora/3.0b2
MIME-Version: 1.0
To: "Diego Caviglia (GA/ERI)" <diego.caviglia@ericsson.com>, Daniele Ceccarelli <daniele.ceccarelli@ericsson.com>, dan.li@huawei.com, "Bardalai, Snigdho" <Snigdho.Bardalai@us.fujitsu.com>, Shan.Zhu@us.fujitsu.com, Igor Bryskin <IBryskin@advaoptical.com>
CC: ccamp@ops.ietf.org
Subject: Re: Working group last call: draft-ietf-ccamp-pc-spc-rsvpte-ext-02
Content-Type: text/plain; charset="ISO-8859-1"; format="flowed"
Content-Transfer-Encoding: 7bit
Authors, I have some LC comments. - First a general comment: I think the document could be significantly improved by centralizing the definition of processing to a new "Procedures" section, rather than the current approach of scattering the definitions across multiple "illustrative" sections. I know this is a significant editing change, but it would result in a extension that is easier to implement, subject to less misinterpretation and less likely to have holes. I see this as a desirable change, but not required. - As always, please cleanup any idnits identified issues. In particular note: == No 'Intended status' indicated for this document; assuming Proposed Standard - The document needs to be run through a spell checker, there are a number of spelling errors. - A general style comment, for consistency I think the document should use the following message names (note the lower cased "message"): OLD NEW "Path Message" "Path message" "Resv Message" "Resv message" "Path Tear Message" "PathTear message" "Path Error Message" "PathErr message" Note if you choose to spell out PathTear message, per rfc2205, it's "Path Teardown". - General comment on document, you use "data-plane" and "Data Plane". Please pick one. - Short title >Internet-Draft RSVP-TE Ext for LSP adoption-ID October 2008 How does " RSVP-TE Ext for LSP adoption-ID" relate to the title "RSVP-TE Signaling Extension For The Conversion Between Permanent Connections And Soft Permanent Connections In A GMPLS Enabled Transport Network."? Perhaps you can use an alternate short title? Also, I suggest adding some consistency in the document in how you refer to the overall process/function. Among others, you use "handover", "adoption", "conversion". While not a critical change, I think picking one and using it consistently would improve the document. - Abstract > Failure conductions and subsequent roll back are also illustrated ^^^^^^^^^^^ -- this should be "conditions", no? - General comment The document uses: >Handover flag >HANDOVER flag >H flag >H-bit Please pick one and be consistent, perhaps "H bit"? see RFC3471 for an example. Also, given that the title of the document is "Conversion" isn't this really the "C" bit/flag? But that's already taken, so I guess you could align the title by replacing "Conversion" with "Handover" in the tile..) - Section 5: > At the ingress node NMS > initiates the transfer of LSP related information residing within > Management Plane to RSVP-TE records within Control Plane. I can only guess what this sentence is trying to say. Can you rephrase? > The following of the section illustrates in detail the procedure in > cases in success and failures. I think you mean "defines" not "illustrates". That is, unless you're going for an informational RFC... - Section 5.1 > Upon receiving a Path Message, each node SHOULD check the consistence > of the actual Data Plane status of such resource. Say the check goes > OK (failure cases are illustrated in the next sections), then a > RSVP-TE record for the LSP is created associating it to the > corresponding Data Plane state. It is not clear (at least to me) at all what an implementation needs to do to conform to this sentence. Perhaps you mean something like "Upon receiving a Path Message containing an Administrative Status Object with the H bit set, a node SHOULD check if there is local Path state matching the MP to CP Handover request. When no local Path state exists, the node SHOULD confirm that there is existing data plane state that corresponds to the Path message. In the case where such data plane state exists (failure cases are illustrated in the next sections), local Path state SHOULD be installed." Also, why aren't the above "SHOULD"s not "MUST"s? > The node accepts all the LSP > information carried in the Path message (if the node is not the > Ingress LER of the LSP, otherwise the information is sent from > relevant MP entity) and stores it in Path State Block. How an implementation stores state is a local matter and outside the scope of a protocol spec, so this sentence should be dropped. >After propagating handover Path message you mean "After propagating a Path message with the H bit set", right? > This means that any memory about the former MP does "memory about" = state? >If a Confirmation message was requested, then it is generated. Does this refer to ResvConf? If then you can something like "Normal ResvConf processing occurs as normal." So what happens once the ingress receives the Resv message with the H bit set? Does it send a Path message with the H bit clear? Does the H bit stay set for the lifetime of the LSP? - Section 5.2.1.1. > case the network status MUST be immediately rollbacked to the one Please define "network status" (or rephrase)? > the Path message or the Resv message forwarding. In this paragraph can't failures occur during any stage of message processing? e.g., as discussed in 5.1. - Section 5.2.1.2. > consist on the unreachability of a node of the LSP. "unreachability", "of the LSP"??? do you mean "inability to reach a node along the path of the LSP""? - Section 5.2.2.1. > In case a failure occurs during the Resv Message forwarding, things > are quite different, because after the handover procedure an LSR is > not able to distinguish an LSP created by the Control Plane from an > LSP created by the Management Plane and then moved to the Control > Plane. Why not? The H bit is still set in the corresponding Path State during almost all cases of Resv processing. This does highlight that the case of a "handover" Resv being received at a node that no longer has corresponding path state. I think you should add something about how this case is handled. > Considering to have a reliable message delivery mechanism, I'm not sure what this clause means or adds. Can you rephrase or drop it? Maybe something along the lines of "When a failure occurs in the processing of a Resv message that contains an Administrative Status Object with the H bit set, the node MUST send ..." > Please note that the failure occurred after the handover procedure > was successfully completed in LSR B. This means that LSR B is no > longer able to know if the LSP was created by the Control Plane or a > handover procedure took place. Same point as above, it will know due to H bit being set in the corresponding Path (and Resv) state. > Upon receiving a Path Tear message, LSR B would delete the LSP also > from the Data Plane point of view. To prevent this from happening, > LSR A MUST set the handover flag in the Path Tear message. I see in reading section 10.2 that this is a different and new "handover flag" carried in the Error Spec Object. (It would have been really useful if you made this clear somewhere more than just buried in section 10.2). In any case this bit isn't really needed as , the downstream node will already have this information due to the H bing having been set in the corresponding Path message/state. So I propose dropping the Error Spec Object "handover flag". > The > downstream node move the LSP ownership back to the Management Plane > and MUST NOT modify the Data Plane. What happens on the upstream transit and ingress nodes? This needs to be specified. - Section 5.2.2.2. > ...Path Error message (with the > Path_State_Removed flag set) Is the setting of the flag a "MUST" or a "SHOULD" in the prior section it was a "SHOULD". > Please note that the failure occurred after the handover procedure > was successfully completed in the Egress LER. This means that it is > no longer able to know if the LSP was created by the Control Plane or > a handover procedure took place. > > Upon receiving a Path Tear message, the downstream nodes would delete > the LSP also from the Data Plane point of view. To prevent this from > happening, the Path Tear message MUST include the Handover flag. > Same 2 comments as previous section. (i.e., don't need new error spec object H bit, as H bit set in corresponding Path message/state.) > ... Due to this > fact, a configurable timer MUST be set on the Ingress LER after > sending the path and on LSR A after forwanding it. When the timer > expires, if no Resv or Path Error message is received, the handover > procedure MUST be aborted and the LSP ownership returned to the > Management Plane. > I understand the timer on the ingress node, but setting a timer on all transit nodes seems to be a pretty major departure from common/standard RSVP processing. For instance, why is such a timer set on normal Path forwarding. I think the timer on the transit nodes really must be dropped. Also, the document needs to specify the processing rules for what happens when the timer expires (on the ingress) and perhaps when the timer should be cleared too. > After sending the path message, the ingress LER sets a timer. If non > Resv message is received before the timer expires, then the adoption > procedure is aborted and the Ingress LER MAY signal it by a Path Tear > message with the H bit set. The document needs to define how standard soft state processing, in particular the handling of timeouts, is impacted by this document. - Section 5.2.2.3. > not release the data-plane resources because in both nodes the H-bit > is set in the PSB. I believe you mean "local Path state" rather than "PSB", correct? > data-plane resources because the H-bit is set in the PSB. Same comment. So does this section impact existing recovery procedures in any way? It seems so, at least in the "release of data-plane resources". If procedures are impacted, then there needs to be some formal RFC2119 language defining the impacted processing. - Section 6.1: > is no more under control of RSVP-TE, which is no more able to "see" I suggest using the word "longer" in place of "more" in this sentence. > This Section covers the procedure needed to manage > this procedure as a dual, opposite procedure respect to the one > described in previous section. This is really not to coherent a sentence, can you rephrase? > Ingress node MUST send out a Path message, with Handover and Reflect > bits in Admin Status set. No action is taken over Data Plane and > Control Plane keeps track of special handover state the LSP is in. > Transit and Egress nodes, upon reception of such handover Path, > propagate it without any Data Plane action, retaining the handover > state information associated to the LSP. After that, every node > waits until the Handover bit is received back in the Resv. Then a > PathTear is issued and the whole LSP information record is cleared > from RSVP- TE data structures. In other words, a normal LSP tear > down signaling is exchanged between nodes traversed by the LSP, but > handover flag set in Path message indicates that no Data Plane action > has to correspond to Control Plane signaling. At the end of handover > tear down signaling flow, the LSP is released from Control Plane > point of view, but its Data Plane state is still unmodified and it is > now owned and controllable by MP. Two points: 1) This section really needs a full rewrite to formally state what are the required ("MUST") and recommended ("SHOULD") actions for each node type (ingress/transit/egress). 2) I think it needs to be clear that the action described in this paragraph are only taken when both the H and D bits are set in the Admin Status Object. (Thanks to Igor for clarifying this for me.) - Section 7. So sections 7 and 8 provide duplicate functionality. Section 7 is actually a fairly major extension in its own right and could be used/specified as a standalone function. The Section 7 function is really an extension to what is already largely present in RFC2745. Rather than propose a fairly major change to this section (and consequently delaying the whole document) I propose that Section 7 be dropped from this document and that if the Authors want to pursue this functionality further it be done so in a separate / stand-alone document. Is this acceptable? I'll deffer detailed comments on this section assuming it is. - Section 8 > downstream label of this interface and the incoming interface ID of > egress node. Why isn't egress node ID sufficient? > Re-iterating this procedure for each transit node along the I suggest replacing "Re-iterating" with "By repeating" > the consistence of resource in Data Plane down to the port level or I suggest replacing "consistence" with "consistency". It's probably worth a note to cover the case where the data plane path continues beyond the egress and that this can be reflected in signaling state using RFC4003/3473. - Section 10.1 This section should not use RFC2119 language for the existing bits, unless those bits are being modified by this document. So the definition of the existing bits, should simply refer back to the existing documents. keep: > - Reflect (R): 1 bit - Defined in [RFC3471] drop: > When set, indicates that the edge node SHOULD reflect the object/ > TLV back in the appropriate message. > keep: > > - Lockout (L): 1 bit - Defined in [RFC4872] drop: > When set, forces the recovery LSP to be temporarily unavailable to > transport traffic (either normal or extra traffic). > keep: > - Inhibit Alarm Communication (I): 1 bit - Defined in [RFC4783] drop: > When set, indicates that alarm communication is disabled for the > LSP and that nodes SHOULD NOT add local alarm information. > > - Call Control (C): 1 bit - Defined in > draft-ietf-ccamp-gmpls-rsvp-te-call-04 [2] drop: > This bit is set when the message is being used to control and > manage a Call. > keep: > - Testing (T): 1 bit - Defined in [RFC3471] drop: > When set, indicates that the local actions related to the > "testing" mode should be taken. > keep: > - Administratively down (A): 1 bit - Defined in [RFC3471] drop: > When set, indicates that the local actions related to the > "administratively down" state should be taken. > keep: > - Deletion in progress (D): 1 bit - Defined in [RFC3471] drop: > When set, indicates that that the local actions related to LSP > teardown should be taken. > > The H bit must be used in conjunction with the R flag when is set in why isn't this a "MUST"? - Section 10.2 > It is possible that a failure, such as the lost of DCN connection or OLD "lost" --> NEW "loss" > In this case the LSP adoption MUST be interrupted and the ownership > of the LSP MUST be moved back to the Plane it belonged to. It is > important that the transaction failure MUST not affect the Data > Plane. The LSP MUST be kept in place and no traffic hit MUST occur. These directives seem to belong elsewhere in the document. If you're simply restating the expected outcome of previously defined procedures, "must" or even "will" should be used rather than "MUST". > messages include an Error_Spec_Object (the Path_State_Removed flag > SHOUL always be set) specifying the causes of the failure, while the Duplicate (and misspelled) directive. > This memo introduces a new Flag and a new Error Code (with different > Error Values) into the Error_Spec Object, defined in [RFC2205]. As discussed above under sections 5.2.2.1/2 I think the HandOverFailure flag is redundant and should be dropped. I'll deffer related detailed comments assuming it is. That's it. Lou On 4/17/2009 12:25 PM, Lou Berger wrote: > > This email begins a two week working group last call on > draft-ietf-ccamp-pc-spc-rsvpte-ext-02.txt > > http://tools.ietf.org/html/draft-ietf-ccamp-pc-spc-rsvpte-ext-02 > > Please send your comments to the list or the authors before the last > call closes on May 1, 2009. > > > >