Hi Jakob, Oh, sorry for the misunderstanding.
So you advised we consider how to handle the label allocation exceptions. Compared to the long duration of label usage, the allocation happens instantly. Once the label is allocated, PEs should keep using it even the network is partitioned. Let's distinguish the exceptions happen in label usage and label allocation. Exceptions on label allocation must be rare. For these rare cases, operators are responsible to fix the label allocation exceptions. >From the side of the network devices, PE should be clear which VPNs should be >protected at the first place. When the primary PE allocates the sharing label >for each VPN, the backup should acknowledge it. If the allocation fails, an >exception should be reported. [Primary after backup] If the backup PE comes online first and already used some labels without considering the protection. Afterwards, the primary PE comes online and operators must have them work in redundant mode. Then operators should be ready to modify labels allocated by the backup to achieve the label sharing proposal. Let's consider another possibility, suppose operators intentionally need the "backup" PE be online first and the "primary" be online later. The "backup" can locally assign a label from the sharing block. When the "primary" comes online afterwards, it should reuse the label that the "backup" used. [Network is partitioned so backup cannot reach primary] Suppose operators have the primary and backup online at the same time. The primary and backup are responsible to establish the ICCP connection with each other. If the primary/backup cannot be reached, an exception should be reported. Please correct me if I'm wrong. Thanks, Mingui >-----Original Message----- >From: Jakob Heitz [mailto:[email protected]] >Sent: Tuesday, November 26, 2013 1:13 AM >To: Mingui Zhang >Cc: UTTARO, JAMES; Lizhenbin; [email protected]; Stewart Bryant (stbryant) >Subject: Re: Why we consider the method of "label sharing for fast PE >protection" > >You missed my point. I asked about a single label. You answered a label block >question. > >-- >Jakob Heitz. > > >On Nov 24, 2013, at 11:41 PM, "Mingui Zhang" <[email protected]> >wrote: > >> Hi Jim & Jakob, >> >> I think you proposed a valid point. If we continue this proposal, we should >include an operational section. I believe this point applies to all such kind >of "fast >egress PE node protection" schemes. >> >>> What if the primary PE comes online after the backup? >>> What label should the backup use if it can not reach the primary? >>> >>> You need to think about partitioned networks too. >> >> If two PEs can reserved a label space to be used as the label range for >> sharing, >the proposal is workable. However, we can slightly relax this constraint, >i.e., PEs >in an RG MUST reserve one label space for sharing. The PEs in an RG just need a >common available label range, right? There are various ways to achieve this. >For >example, we can divide the label spaces into several blocks beforehand. Give >them priorities. When PEs locally allocate labels, blocks with the higher >priority >will be used earlier. Then PEs in an RG can agree on those common blocks with >lower priorities. This relaxation introduces flexibility. Imagine the case >that an >operator may be not sure about the size of the label space to be reserved at >the >beginning. In this way, even the primary come online after the backup, it can >negotiate a label range for sharing with the backup. >> >> At last but not least, even if PEs cannot find a common label range in the >> worst >case or the network is partitioned, the ICCP connection will not be >established. >In this case, the PEs just fall back to the non redundant mode. So the two >scenarios proposed by Jakob can be handled. >> >> Anyway, I don't think there is no way to achieve such proposal. >> >> Thanks, >> Mingui >> >>> -----Original Message----- >>> From: L3VPN [mailto:[email protected]] On Behalf Of UTTARO, >>> JAMES >>> Sent: Friday, November 22, 2013 4:17 AM >>> To: Jakob Heitz; Lizhenbin >>> Cc: [email protected]; Stewart Bryant (stbryant) >>> Subject: RE: Why we consider the method of "label sharing for fast PE >>> protection" >>> >>> As I stated before an operational section that anticipates running >>> the network would be useful in this discussion. I would like to see >>> various scenarios, restrictions, failure modes etc.. >>> >>> Jim Uttaro >>> >>> -----Original Message----- >>> From: Jakob Heitz [mailto:[email protected]] >>> Sent: Thursday, November 21, 2013 3:09 PM >>> To: Lizhenbin; UTTARO, JAMES >>> Cc: [email protected]; Stewart Bryant (stbryant) >>> Subject: RE: Why we consider the method of "label sharing for fast PE >>> protection" >>> >>> What if the primary PE comes online after the backup? >>> What label should the backup use if it can not reach the primary? >>> >>> You need to think about partitioned networks too. >>> It's not as simple as you think. >>> >>> -- Jakob. >>> >>> -----Original Message----- >>> From: Lizhenbin [mailto:[email protected]] >>> >>> It is not like this way. For the ICCP-based label allocation method, >>> the primary PE is allocate the label for VPN by its own label >>> allocation algorithm. The backup PE is just to allocate the label >>> specified by the primary PE. The process is just like label backup. They >>> need >not use the same label allocation algorithm.
