|
Hi, That is a good start. The general problem for this type of problem is that one can reasonably calculate a pacing schedule based on target bit-rate at the outgoing interface.
What one doesn’t know is if what path the various message takes and if that is part of a traffic load causing congestion. The Map-Server will get some indication on potential congestion issue if it has to retransmit many messages as they aren’t acked. I would
think the most general thing I would say is to recommend that the pacing target a bit-rate that is no more than a small fraction of the expected bandwidth of the links to the xTRs.
That is likely preventing enough issues that there is no point in doing more advanced solution. But that is me assuming that the control plane will mostly run over links
with Gbps+ capacity and that one configure this to not burst above like 1-5% of the link capacity one will not have any issues. But if there is more limited capacity or larger deployments maybe the completion time become an issue for each update. Cheers Magnus From:
mohamed.boucadair@xxxxxxxxxx <mohamed.boucadair@xxxxxxxxxx> Re-, Thanks Magnus for clarifying.
I suggest to add the following in Section 6: NEW: As a reminder, the initial transmission and retransmission of Map- Notify messages by a Map-Server follow the procedure specified in Section 5.7 of [RFC9301]. Some state changes may trigger an overload that would impact, e.g., the outbound capacity of a Map-Server. A similar problem may be experienced when a large number of state were simultaneously updated. To prevent such phenomena, Map-Servers SHOULD be configured with policies to control the maximum number of subscriptions and also the pace of Map-Notify messages. The exact details to characterize such policies are deployment and implementation specific. Likewise, this document does not specify which notifications take precedence when these policies are enforced. Do we need to say more without going too much into implementation territory? Cheers, Med De : Magnus Westerlund <magnus.westerlund@xxxxxxxxxxxx>
Hi Med,
MW: Yes, the issue is that in the context of subscriber service, I think it is relevant to discuss how to ensure that the server does not try to overload its own outgoing
paths as N subscribers to a particular mapping will result in N outgoing message when that mapping is updated. That N messages can’t in general be sent all immediately as it will result in a large spike, for large enough values of N overloading. That needs
some discussion. And as I discussed in my review the retransmission timer being fixed at 3 seconds will need to be taken into account. Because if one pace so that the pacing of the N Map-Notify so that it will take more than 3 seconds then the retansmissions
will also start result in additional load when they occur.
My point is that the pubsub mechanism creates new issues as it both build up states and when the Map-Notify messages are triggered the whole state needs to be considered
in doing reasonable things. Cheers Magnus _________________________________________________________________________________________________________________________
Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
This message and its attachments may contain confidential or privileged information that may be protected by law;
they should not be distributed, used or copied without authorisation.
If you have received this email in error, please notify the sender and delete this message and its attachments.
As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
Thank you.
|
-- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
