I was wondering if anyone has experience using MT as a transparent bridge doing QoS (PCQ, DSCP, marks, etc . . .)/prioritization on a large scale, i.e. thousands of users and 300-400Mbps+ of traffic. Possible with the appropriate x86 hardware? Just to throw a number out there say 50-100 mangles.
Well, no takers, lol. I’ll post results when I have them, in case anyone else is interested.
Bridging uses CPU from what I recall so your going to need some grunt, Until ROS 5 is stable the fastest single core cpu you can find along with well supported gig nic’s should do it
Fast hardware I have.
the main question is: does anybody have an experience with QoS on hundreds Mbps? =)
Some results from initial testing, starting out small.
Bandwidth manager is a Supermicro P4 3.0Ghz system with two Intel GigE cards.
83 mangle rules that mark and prioritize traffic, includes use of MT all-p2p filter. 1 queue tree with 3 pcq sub-queues managing bandwidth in the download (toward end users) direction.
Max incoming traffic from the Internet is approximately 65Mbps, large variety of traffic (P2P and streaming galore) generated by students. Incoming packet rate around 9,000 PPS. Approximately 76k connections. At these rates, I see a max of approximately 25% CPU usage; averaging in the 15-20% range.
Not too shabby. I have a tricked out X3460 system that will be the ultimate platform. We are creating a network testing tool to approximate real-world usage scenarios at very high speeds.