you know the situation. You start something, then your network scales, it mostly works, so you have no need to improve the situation. However, we want to change few things, so I have following questions:
- currently we use simply queues only for our clients. One consultant told us, that SQ does not equaly give connect to your clients. E.g. you have 2mbit line, 10 512kbit users in theory. Then first 4 are satisfied only. So, if he is true, how to start lowering the connection for each user equally? We currently use default queue type, which uses RED.
- how is that I can't define parent for simple queue? We tried it, but sum of child queues exceeded max-limit of parent. Is SQ buggy?
- currently we don't use any means of guaranteeing our subnodes certain traffic. I thought that simply each router takes what is needed Is it wise to precisely divide your main router connection between subsequent nodes? Is it wise for the router to know what it receives? Does it influence/improves algorithm of queueing/shaping?
- where to limit users? First node they connect to, or the main router?
- I know some ppl prefer queue tree. But we already saw one big provider, who gave up on QT, because for more than 700 users there were so many entries in QT, that it started to kill router. That is strange, no? I thought that SQ in fact defines three queues in the background anyway?
- as in other thread - we are starting to consider two separate routers - one for NAT, second for queueing/shaping - will it in fact help? I mean - e.g. p2p marking can't match everything, and my understanding is, that NAT could be the culprit?
Any advice appreciated
Thanks a lot,