Hi PANRG members: Thank you for all the valuable questions and feedback provided during our PANRG talk last week about computing path metrics using bottleneck structures graphs (BSGs, https://datatracker.ietf.org/meeting/115/materials/slides-115-panrg-bottleneck-structure-graphs-in-multidomain-networks-introduction-and-requirements-for-alto-draft-giraltyellamraju-alto-bsg-multi). We hope we can initiate a discussion to collect feedback on how this work could be useful to the goals of PANRG. To summarize, in our talk:
* We explained how BSGs can be used to compute optimized traffic engineering solutions. For instance, the bottleneck structure provides a metric for each path that characterizes the expected throughput of a flow that's placed on that path. Use cases of BSGs are numerous, including routing, congestion control, service placement, network slicing, SLA management, capacity planning, among others. * We provided a distributed border protocol to compute BSGs under partial information (multiple autonomous systems case). The protocol is proven to converge to the correct global BSG using local AS information and requiring only that each AS shares its path metric dictionary with its neighbor ASs. * A worst case upper bound on convergence time of the protocol is obtained when each AS degenerates to a single switch. In this case, the convergence time is equal to the time it takes for a packet to traverse all links in the BSG. This is in general fast (e.g., in the order of 1 second). In reality, since an AS will have more than one single switch, convergence time will be lower than this bound. * The protocol avoids the need to share sensitive information such as topology or flow information between ASs. One of the comments provided during the session by the chair was that PANRG might be interested in discussing the protocol for signaling the path metrics. Another discussion topic that was brought up during the session focused on use cases that can illustrate how BSGs can be used to solve network optimization problems (including traffic engineering). Would these two topics (signaling protocol and use cases) be a good starting point for this discussion? Any other topics you'd like to discuss? I am CC-ing the ALTO WG since its members are also discussing BSGs in the context of the ALTO standard. Thank you. Jordi
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
