1

I have a linux based traffic shaper (iptables + tc htb policy). It works in bridge mode. Shapes traffic based on IPs and ports (there are about 100 rules in the "mangle" chain of iptables). Right now its throughoutput is about 100 mb/s (I don't remember pps, there are about 800 users in the network). Just was wondering - when I will hit the limit. How much traffic could a linux-based shaper possibly get throuhg it. If you have one under heavy load, please could you write what machine you use and what load there is. Or if you have any other info about the subj, please write as well.

Thanks in advance.

facha
  • 1,298
  • 2
  • 16
  • 26
  • don't forget, throughput it important, but think also about latency. Your setup sounds very cool, do you care to share your configurations? – The Unix Janitor Apr 14 '10 at 14:31

2 Answers2

1

You should be able to do a good few hundred megabit. Load balancing's a lot easier so I'd say Chopper3's answer doesn't really apply.

By far the best thing to do is look at the CPU logs in the system stats you should already be recording (Cacti is my preference, but many people swear by Munin) and compare them with the bandwidth graphs. In this case you are primarily concerned with System CPU usage.

It's so machine and kernel dependent that the only real answer is find out for yourself.

It is worth looking at the iptables stats and reorder rules so that more commonly hit rules match as early as possible (creating separate chains for each in/out interface pair can be a massive performance gain in some cases).

LapTop006
  • 6,466
  • 19
  • 26
0

It depends on your server, it's components (CPU/s, NICs, buses, memory speed etc.) and OS/config but we have single servers happily load-balancing 8-9Gbps all day every day.

Chopper3
  • 100,240
  • 9
  • 106
  • 238