Hello, I’d like to test Virtual Network Embedding algorithms (VNE) in the wild for my bachelor thesis. These algorithms find the optimal [1] mapping of a virtual network (e.g. services and relations) onto the actual hardware. However, most algorithms only take the CPU power and bandwidth of the links between machines into account. It would be really interesting to see what happens if there’s latency involved and how to adapt algorithms with that in mind. Juju is the perfect project to implement and test an embedding algorithm. But being quite new to Juju, I have two related questions:
- From the code in state/state.go it seems that Juju will either deploy to the „first“ available machine or start a new one to deploy a service. In other words: the way new services are deployed is not yet optimized. Is that correct? - Is latency between machines a factor when using Juju? Services and relations are strictly organized in environments and so only machines in the same region talk to each other. Are there constellations where latency is an issue? I really appreciate any insight. Of course, if there’s room for improvement, I’ll gladly provide patches. Thanks Pascal [1] The problem is NP-hard, so heuristic algorithms are used. -- Juju mailing list [email protected] Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/juju
