On 04/12/2019 20:46, Harald Welte wrote: > >> We don't envisage a separation between MSC and HLR over unreliable >> back-haul, but I think I'm missing something here. (I still need to >> actually implement a local setup and observe) > In an inbound roaming situation, you have the MSC (VPLMN) in one village and > the > "authoritative" HLR for that subscriber (HPLMN) in another village. > Ah, indeed. I do need to setup and play with this. In my head, I was imagining that an IMSI attach request triggers an mslookup (broadcast) and that the response message contains what the VLR (msc) needs. I wasn't imagining TCP connections relating to core GSM over the unreliable IP network.
I'm wary of TCP. I'm no expert in all the things that can be tuned, so maybe there's a solution I'm unaware of? If not, I don't like tcp over tinc-vpn at all where the underlying IP connection is unreliable. Things tend to stall for a long time. This is precisely the main problem I was seeing with multi-master distributed databases. They work fine in theory in a data centre, but add severe packet loss and it all falls down. I'm quite possibly still missing something? Thanks! k.
