Re: Bandwidth modification in GMPLS
Dimitri Papadimitriou <dimitri.papadimitriou@alcatel.be> Wed, 05 December 2001 11:00 UTC
Envelope-to: ccamp-data@psg.com
Delivery-date: Wed, 05 Dec 2001 03:02:52 -0800
Message-ID: <3C0DFE44.126D64ED@alcatel.be>
Date: Wed, 05 Dec 2001 12:00:20 +0100
From: Dimitri Papadimitriou <dimitri.papadimitriou@alcatel.be>
Organization: Alcatel Bell - IPO NA (NSG)
MIME-Version: 1.0
To: Maarten Vissers <mvissers@lucent.com>
CC: manoj juneja <manojkumarjuneja@hotmail.com>, ccamp@ops.ietf.org, kireeti@juniper.net, eric.mannie@gtsgroup.com, "Helvoort, Huub van" <hhelvoort@lucent.com>
Subject: Re: Bandwidth modification in GMPLS
Content-Type: multipart/mixed; boundary="------------1881898109D5420C5A141631"
Maarten, See in-line... Maarten Vissers wrote: > > Dimitri, > > I am not comparing the two methods. I am describing how LCAS operates the SEQ > IDs and how connection management has to support virtual concatenation with > LCAS. GMPLS is an implementation of distributed connection management, and must > therefore support what I described. This is what i was trying to demonstrate (see previous email). > In addition, connection management must support a means to perform the same > capabilities for the case one or both endpoints do not support LCAS, but allow > resizing the pipe. The proposal you made in your LBM document is much more > restrictive than LCAS capable implementations. It is up to the operators to > decide if these restrictions are acceptable. It is most likely a trade of > between less restrictions/more complex protocol and more restrictions/less > complex protocol. This is a first proposal be sure the next version will be improved. > Personally, I would like to compare two proposals: one that supports all LCAS > capabilities (except of course the hitless one), the other that includes > restrictions. If the complexity difference is not too large, go for the full > capabilities support, otherwise select the one with the restrictions. Agreed. But one should be in agreement in what we consider by "not too large" Regards, - dimitri. > Regards, > > Maarten > > Dimitri Papadimitriou wrote: > > > > Maarten, > > > > First for people reading these e-mails the discussion here > > is not focused on a comparison between LCAS versus something > > else but only on control plane issues (more precisely which > > are the GMPLS mechanisms to be provided in order to enable > > bandwidth modification on different scenarios) > > > > Now, to continue the discussion we have defined in the GMPLS > > LBM I-D the Circuit ID and the LSP Sequence Number for that > > purpose (see Section 5.2). Together with the following rule: > > > > "All LSPs belonging to the same circuit MUST have the same > > Circuit ID. All LSPs MUST have a unique LSP sequence number > > in that Circuit ID, the LSP sequence number MUST be the > > sequence number of the first SPE/VC of the overall virtually > > concatenated circuit." > > > > Therefore, the scenarios you have defined are a combination > > of the methods described in Section 5.1 and 5.2. > > > > To illustrate this, let's take your example back: > > > > > Assume the connection controller selects VC-4 SNPs with addresses G1, G2 to > > > connect to route A, and G3, G4 to connect to route B. G5 to G7 are not connected > > > VC-4 SNPs. > > > > > > VC-4 with G1 via route A > > > VC-4 with G2 via route A > > > VC-4 with G3 via route B > > > VC-4 with G4 via route B > > > VC-4 with G5 not connected > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > => Circuit ID #1 (for instance) will include LSP SEQ ID#0 > > and LSP SEQ ID#1 (here we assume that Route A is link > > diverse from Route B to preserve generality of the > > proposed method) > > > > > Assume that the VC-4 trail via G2 is the first one which clears its Signal Fail > > > (SF) condition, then this trail will pass through the signal with SEQ ID=0. If > > > the second one is via G4, then this one passes through the signal with SEQ ID=1. > > > If 3rd is via G3, then this gets SEQ ID=2 and finally via G1 gets SEQ ID=3. > > > > > > VC-4 with G1 via route A SEQ ID=3 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route B SEQ ID=2 > > > VC-4 with G4 via route B SEQ ID=1 > > > VC-4 with G5 not connected > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > => As such LSP SEQ ID#0 corresponds to the SEQ ID=0 and LSP > > SEQ ID#1 to the SEQ ID=0 > > > > > When adding the 5th VC-4 (e.g. via VC-4 SNP G5), this VC-4 will carry SEQ ID #4. > > > > > > VC-4 with G1 via route A SEQ ID=3 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route B SEQ ID=2 > > > VC-4 with G4 via route B SEQ ID=1 > > > VC-4 with G5 via route A SEQ ID=4 > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > => To update the LSP SEQ#0 you follow the rules defined in > > the Section 5.1 > > > > > Now route B fails; i.e. VC-4 with SEQ ID #1 and #2 fail. There will be no change > > > of SEQ ID, only there will be an indication not to transmit over those VC-4 > > > trails. If later on it is decided to continue over route A only, the SEQ IDs > > > are adapted as follows: > > > > > > VC-4 with G1 via route A SEQ ID=1 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 not connected > > > VC-4 with G4 not connected > > > VC-4 with G5 via route A SEQ ID=2 > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > => you delete the VC-4 components with the re-sequencing as described > > at the end of the Section 5.1 with the restriction that "SPEs/VCs > > can be removed from an LSP but only at the end of that LSP (i.e. > > starting at SPE/VC sequence number X)" > > > > > If the demand for bandwidth doubles again, it is possible to add say three more > > > VC-4 trails via e.g. route C (X - C - Y); e.g. via G3, G4 and G6. If the order > > > those trails become available is G6, G3, G4 then the SEQ IDs will be: > > > > > > VC-4 with G1 via route A SEQ ID=1 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route C SEQ ID=4 > > > VC-4 with G4 via route C SEQ ID=5 > > > VC-4 with G5 via route A SEQ ID=2 > > > VC-4 with G6 via route C SEQ ID=3 > > > VC-4 with G7 not connected > > > > => once again you follow the rule described in Section 5.1 > > > > There is no need to use the multiplier for this purpose (this > > field is used for instance to operate a change on multiple LSP > > using the same message - it has been provided to avoid lengthy > > repetitions of the same operation) > > > > Once again (i think that Eric has already provided this comment) > > the scope of the LSP SEQ ID is not to reproduce the LCAS defined > > structures within the control plan but to provide a consistent > > control plane definition (allowing a level of consistency checking > > between the control plane and the transport plane.) > > > > Are we in agreement, or do you want further comparison of both > > methods ? > > > > Regards, > > - dimitri. > > > > Maarten Vissers wrote: > > > > > > Dimitri, > > > > > > Assume the VC-4-4v signal consisting of 4 VC-4 signals. Each VC-4 signal has a > > > SEQ ID (0,1,2,3). Which SEQ ID is associated with which VC-4 is not not fixed... > > > this is under control of LCAS and may change with time... > > > > > > When you set up the connection to transport the VC-4-4v signal via two routes > > > (A, B), two VC-4 termination points will be connected to the A route, two other > > > to the B route. > > > There are typically 7 VC-4 termination points per GbE port... let's refer to > > > those 7 VC-4 termination points by means of SubNetwork Point (SNP) addresses G1 > > > .. G7. > > > > > > Assume the connection controller selects VC-4 SNPs with addresses G1, G2 to > > > connect to route A, and G3, G4 to connect to route B. G5 to G7 are not connected > > > VC-4 SNPs. > > > > > > VC-4 with G1 via route A > > > VC-4 with G2 via route A > > > VC-4 with G3 via route B > > > VC-4 with G4 via route B > > > VC-4 with G5 not connected > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > > > Assume that the VC-4 trail via G2 is the first one which clears its Signal Fail > > > (SF) condition, then this trail will pass through the signal with SEQ ID=0. If > > > the second one is via G4, then this one passes through the signal with SEQ ID=1. > > > If 3rd is via G3, then this gets SEQ ID=2 and finally via G1 gets SEQ ID=3. > > > > > > VC-4 with G1 via route A SEQ ID=3 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route B SEQ ID=2 > > > VC-4 with G4 via route B SEQ ID=1 > > > VC-4 with G5 not connected > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > > > When adding the 5th VC-4 (e.g. via VC-4 SNP G5), this VC-4 will carry SEQ ID #4. > > > > > > VC-4 with G1 via route A SEQ ID=3 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route B SEQ ID=2 > > > VC-4 with G4 via route B SEQ ID=1 > > > VC-4 with G5 via route A SEQ ID=4 > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > > > Now route B fails; i.e. VC-4 with SEQ ID #1 and #2 fail. There will be no change > > > of SEQ ID, only there will be an indication not to transmit over those VC-4 > > > trails. If later on it is decided to continue over route A only, the SEQ IDs > > > are adapted as follows: > > > > > > VC-4 with G1 via route A SEQ ID=1 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 not connected > > > VC-4 with G4 not connected > > > VC-4 with G5 via route A SEQ ID=2 > > > VC-4 with G6 not connected > > > VC-4 with G7 not connected > > > > > > If the demand for bandwidth doubles again, it is possible to add say three more > > > VC-4 trails via e.g. route C (X - C - Y); e.g. via G3, G4 and G6. If the order > > > those trails become available is G6, G3, G4 then the SEQ IDs will be: > > > > > > VC-4 with G1 via route A SEQ ID=1 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route C SEQ ID=4 > > > VC-4 with G4 via route C SEQ ID=5 > > > VC-4 with G5 via route A SEQ ID=2 > > > VC-4 with G6 via route C SEQ ID=3 > > > VC-4 with G7 not connected > > > > > > Looking at the connection setup details, I do not setup a VC-4-2v via A and > > > another VC-4-2v via B initially... I do setup two groups of two VC-4 > > > connections, which together support the VC-4-4v trail; a VC-4-4v is not the > > > same as two times a VC-4-2v. The use of NVC parameter here seems as such not > > > appropriate. The only alternative present is the MT field. > > > > > > Alternative is to redefine NVC as "This field indicates the number of identical > > > co-routed connections that are requested for this LSP. These connections are all > > > of the same type." > > > > > > But if I assume for a moment an alternative example in which the > > > virtual concatenated endpoints are outside of the network of > > > interest, then a UNI/E-NNI request using MT will result in the > > > same connections being setup compared to such request using the > > > modified NVC definition. What am I missing here? > > > > > > The traffic parameters for these two group-connections supporting the VC-4-Xv > > > trail are thus: > > > > > > initially: > > > route A: ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 [or NVC=2, MT=1] > > > route B: ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 [or NVC=2, MT=1] > > > > > > after 5th VC-4 trail addition: > > > route A: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 [or NVC=3, MT=1] > > > route B: ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 [or NVC=2, MT=1] > > > > > > after route B connections being released: > > > route A: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 [or NVC=3, MT=1] > > > > > > after 4th, 5th, 6th connection being added via route C: > > > route A: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 [or NVC=3, MT=1] > > > route C: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 [or NVC=3, MT=1] > > > > > > To add the 3rd VC-4 connection to the VC-4 group-connection over route A, a > > > MODIFY message is required. In your LBM document you propose to signal the diff > > > value only, and consequently let the switch nodes compute the new total (i.e. > > > new = original + diff). It will work as well to signal the new value and let the > > > switch nodes compute the difference (i.e. diff = new - original). We need to > > > evaluate the pros/cons of both approaches before selecting one. > > > > > > When the group-connection via route B is removed, this group-connection has to > > > be released. > > > > > > When the group-connection via route C is to be added, this group-connection has > > > to be setup. > > > > > > In addition to these cases, the MODIFY message has to support also the decrease > > > case. E.g. assume that route A is more expensive than route C, then one would > > > obviously prefer to release a VC-4 connection via route A when the total > > > bandwidth of the VC-4-Xv has to be shrunk from 900 Mb/s to 750 Mbit/s. > > > > > > after 3rd connection via route A is removed: > > > route A: ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 [or NVC=2, MT=1] > > > route C: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 [or NVC=3, MT=1] > > > > > > VC-4 with G1 via route A SEQ ID=1 > > > VC-4 with G2 via route A SEQ ID=0 > > > VC-4 with G3 via route C SEQ ID=3 > > > VC-4 with G4 via route C SEQ ID=4 > > > VC-4 with G5 not connected > > > VC-4 with G6 via route C SEQ ID=2 > > > VC-4 with G7 not connected > > > > > > When using a MODIFY message which includes the new traffic parameters (and not > > > the diff ones), the same MODIFY message can be used for addition and removal of > > > connections to/from the group. This looks like a pro for a MODIFY message > > > definition including the new traffic parameters. > > > > > > One note to the above scenarios... the set of 7 VC-4 SNPs can be operated as a > > > single SNP Pool (SNPP). Instead of requesting the setup of a group-connection to > > > SNPs G1 and G2 (route A) and to SNPs G3 and G4 (route B), one could choose to > > > request the setup of two 2-group-connections to SNPP "G1-G7" and let the local > > > switch node decide which specific SNPs will be allocated. > > > And the same SNPP "G1-G7" address can be used to add the 5th VC-4 connection, > > > and to add the 4/5/6th VC-4 connections via route C later on. > > > > > > I hope this clarifies some of the capabilities of the control plane needs to > > > support when a group-connection is to be modified, and the operation of the SEQ > > > ID within virtual concatenation with LCAS. > > > > > > Regards, > > > > > > Maarten > > > > > > Dimitri Papadimitriou wrote: > > > > > > > > Maarten, > > > > > > > > As per GMPLS-LBM, the multiplier field (MT) SHOULD not be used > > > > for bandwidth increase purpose since it is used for a different > > > > semantic. > > > > > > > > SPEs/VCs can only be added at the end of an LSP. For instance, > > > > when adding Y VC-4s to a VC-4-Xv (where each of the VC-4 is > > > > numbered from 0 to X-1), one obtains a VC-4-(X+Y)v with the > > > > sequence numbered from 0 to X+Y-1. > > > > > > > > The number Y of SPEs/VCs to add is given by the following > > > > equation: new NVC - old NVC. The first new SPE/VC will have a > > > > sequence number equals to the sequence number of the previously > > > > last VC + 1 (i.e. X). > > > > > > > > In such a case (in your below example) > > > > > > > > - Connection X - A - Y traffic parameters > > > > * Before increase VC-4-2v: ST=6, RCC=0, NCC=0, NVC=2, MT=1, T=0 > > > > * Increase to VC-4-3v: ST=6, RCC=0, NCC=0, NVC=1, MT=1, T=0 > > > > > > > > remember that MT = 0 is an invalid value (as per draft-ietf-ccamp > > > > -gmpls-sonet-sdh-02.txt) > > > > > > > > The MT field in this picture is used in the following way (the > > > > number Y of SPEs/VCs to add is given by the following equation: > > > > (new NVC * new MT) - (old NVC * old MT) just notice that in most > > > > cases new MT will be equal to old MT): > > > > > > > > - 2 Connections X - A - Y traffic paramters > > > > * Before increase VC-4-2v: ST=6, RCC=0, NCC=0, NVC=2, MT=2, T=0 > > > > * Increase to VC-4-3v: ST=6, RCC=0, NCC=0, NVC=1, MT=2, T=0 > > > > > > > > Your second example is depicted in the section 5.2 of the > > > > draft-mannie-ccamp-gmpls-lbm-tdm-00.txt but again the use > > > > of the multiplier (MT) in that context is not recommended. > > > > > > > > Regards, > > > > - dimitri. > > > > > > > > Maarten Vissers wrote: > > > > > > > > > > Manoj, > > > > > > > > > > There is a difference between > > > > > 1) modifying the bandwidth of a connection and > > > > > 2) replacing the connection by a different connection. > > > > > > > > > > Add 2) A VC-3 connection is a connection in a VC-3 network layer, whereas a VC-4 > > > > > connection is a connection in a VC-4 network layer. These network layers are > > > > > independent, and as such you can only replace a VC-3 connection by a VC-4 > > > > > connection by releasing the VC-3 connection and requesting to set up a VC-4 > > > > > connection. > > > > > > > > > > Add 1) You can modify the "bandwidth" of a group-connection (i.e. NVC(*) or MT > > > > > >= 1), by adding/removing connection members. If this group-connection supports a virtual concatenated trail (VC-n-Xv, ODUk-Xv) then the supported client link's bandwidth is modified as a result. > > > > > > > > > > Example: Consider VC-4-Xv trail between nodes X and Y which > > > > > supports an Ethernet link between these two nodes. When X is > > > > > increased from 4 to 5, the Ethernet link bandwdith is increased > > > > > from 600 Mb/s to 750 Mbit/s. > > > > > > > > > > Assume that the VC-4-Xv trail is supported via two diverse > > > > > routes (X - A - Y and X - B - Y), each providing two VC-4 > > > > > connections. Then the bandwidth modification will increase > > > > > the number of connections on one of the two routes; e.g. the > > > > > route via A. The associated traffic parameters will be: > > > > > > > > > > - Connection X - A - Y traffic parameters > > > > > * Before increase: ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 > > > > > * After increase: ST=6, RCC=0, NCC=0, NVC=0, MT=3, T=0 > > > > > - Connection X - B - Y traffic parameters (unchanged) > > > > > * ST=6, RCC=0, NCC=0, NVC=0, MT=2, T=0 > > > > > > > > > > What is needed is a MODIFY request which allows for an identified connection to > > > > > modify the MT value. > > > > > > > > > > (*) Note that the parameter NVC is interpreted as to represent the number of > > > > > components in a virtual concatenated signal connection... in this example NVC=4 > > > > > and grows to NVC=5. As the virtual concatenated signal is transported via two > > > > > routes, each having half of the member connections, I have the impression that > > > > > MT is the parameter to use. > > > > > > > > > > Or to turn this around and look at the source for the bandwidth modification > > > > > first, we may assume that over the e.g. Ethernet UNI the call controller > > > > > (G.8080) at the network side receives a request to increase the enthernet link > > > > > from 600 to 750 Mbit/s. The call controller then decides to support this > > > > > ehternet bandwidth modification request by means of a modification of the > > > > > group-connection via route A... it then issues a connection modification request > > > > > for this group-connection... etc. > > > > > > > > > > Add 1) True bandwidth modification of an individual connection is possible for > > > > > e.g. ATM VP, ATM VC and MPLS connections. As long as the ATM VP, ATM VC or MPLS > > > > > links (G.805) over which the connection is transported have spare capacity, the > > > > > bandwidth of an individual VP, VC or LSP can be increased. > > > > > > > > > > Regards, > > > > > > > > > > Maarten > > > > > > > > > > manoj juneja wrote: > > > > > > > > > > > > Hi All, > > > > > > If the bandwidth modification is not supported by GMPLS then > > > > > > it should be clearly mentioned in the drafts (at least the architecture > > > > > > draft). Please see the attached mail for reference. > > > > > > > > > > > > I am still not clear whether GMPLS supports SE-style of reservation. > > > > > > My original question related to SE-style of reservation was and I didn't > > > > > > get any reply on the mailing list : > > > > > > > > > > > > " I have a doubt related to SE style of reservation in GMPLS. As > > > > > > in RSVP-TE, multiple senders can share the reservation but they can be > > > > > > assigned different labels to distinguish different senders from one > > > > > > other. > > > > > > In GMPLS, as the label itself is a resource on the link, it means that > > > > > > if there are multiple senders in the same session sharing reservation, > > > > > > they should be assigned the same label. They can't be assigned > > > > > > different labels or is it that GMPLS does not support SE style of > > > > > > reservation ?" > > > > > > > > > > > > Please correct me if I am missing something. > > > > > > > > > > > > Regards, > > > > > > manoj. > > > > > > > > > > > > >From: Maarten Vissers <mvissers@lucent.com> > > > > > > >To: manoj juneja <manojkumarjuneja@hotmail.com> > > > > > > >CC: ccamp@ops.ietf.org > > > > > > >Subject: Re: Bandwidth modification in GMPLS > > > > > > >Date: Fri, 30 Nov 2001 11:45:25 +0100 > > > > > > > > > > > > > >Manoj, > > > > > > > > > > > > > >manoj juneja wrote: > > > > > > > > > > > > > > > > Hi All, > > > > > > > > Is it possible to modify/increase the bandwidth of an existing > > > > > > >GMPLS > > > > > > > > tunnel ? I mean to say if I establish a VC-3 SDH LSP, will it be > > > > > > >possible to > > > > > > > > increase the bandwidth to say VC-4 ? > > > > > > > > > > > > > >No, this is not possible. > > > > > > > > > > > > > >Regards, > > > > > > > > > > > > > >Maarten > > > > > > > > > > > > > > > > > > > > > > > Should the new request be considered independent of previous one ? > > > > > > > > > > > > > > > > Regards, > > > > > > > > manoj. > > > > > > > > > > > > > > > > _________________________________________________________________ > > > > > > > > Get your FREE download of MSN Explorer at > > > > > > >http://explorer.msn.com/intl.asp > > > > > > ><< mvissers.vcf >> > > > > > > > > > > > > _________________________________________________________________ > > > > > > Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp
- Re: Bandwidth modification in GMPLS manoj juneja
- Re: Bandwidth modification in GMPLS Maarten Vissers
- Bandwidth modification in GMPLS manoj juneja
- Re: Bandwidth modification in GMPLS Dimitri Papadimitriou
- Re: Bandwidth modification in GMPLS Maarten Vissers
- Re: Bandwidth modification in GMPLS manoj juneja
- Re: Bandwidth modification in GMPLS Dimitri Papadimitriou
- Re: Bandwidth modification in GMPLS Maarten Vissers
- Re: Bandwidth modification in GMPLS Dimitri Papadimitriou
- Re: Bandwidth modification in GMPLS Maarten Vissers