1

Please note that I am not asking for well known usual bandwidth monitoring tools like ntop, iperf, vnstat etc.

I would like to measure periodically and reliably the max. available bandwidth of a network uplink, but without "stealing" bandwidth and without disturbing other traffic on this same connection.

Is this possible at all?

Usually I do adhoc measurements with rsync or iperf, but from what I understand these kind of tests also saturate the bandwidth of the connection and this makes it not possible to do this, say, every minute.

I have a not so stable uplink with several remote desktop connections and every bit counts, but I also would like to collect meaningful logfiles to be able to present information about this connection not offering the contracted bandwith at all.

So how do I collect this data without killing the already weak network connections going on here?

2 Answers 2

2

Is this possible at all?

No, this is not possible. By definition, measuring bandwidth requires actually sending/receiving traffic, and since networking gear has a finite amount of information it can pass in a certain time period, you will be consuming a certain amount of that finite resource.

Sure, you could turn on QoS in your switch/routing gear and de-prioritize your bandwidth testing traffic, but then you'll not be getting accurate statistics on bandwidth available.

So how do I collect this data without killing the already weak network connections going on here?

You do your testing during periods of low load.

The other thing you may consider doing is setting up something like Cacti to monitor and graph bandwidth usage over time - this will give you quite a bit of data to work with.

4
  • 1
    Ah, come on, I expected some little unknown piece of network technicians magic getting revealed... so I will just wait a little bit longer before accepting your answer, but, well, I believe logic is stronger than magic... :) Jan 28, 2015 at 19:55
  • Haha, unfortunately there's no magic to be had here. :)
    – EEAA
    Jan 28, 2015 at 19:56
  • btw I am using ntop already, but it does not give me any information about the max. possible bandwidth. If, by accident, the bandwidth is saturated, well, then of course the ntop graphs allow a well educated guess about what might be the max. bandwidth, but there is no single source of truth about the real max. bandwidth available at any point in time... Jan 28, 2015 at 19:59
  • There is a single source of truth about the max available bandwidth. It involves reading the specifications of every active and passive component in the signal chain. Unfortunately if you do not control things from end-to-end, this can be difficult. Additionally, RF/Optical links add an additional layer of ambiguity. Your best bet is probably to get a provider that can give you an SLA.
    – EEAA
    Jan 28, 2015 at 20:01
-2

Back in the mists of time, I heard about this tool: bing. It has some limits but it may be of help even though, of course, it needs to send some traffic through the links we are trying to determine the effective bandwidth of.

Hope this helps!

3
  • There's also iperf. Neither of these tools do what the OP is asking about, though, because it's impossible.
    – EEAA
    Jan 28, 2015 at 20:17
  • First of all, I am not sure the asked question refers to measuring bandwidth without sending any traffic at all or without sending amounts of traffic that will likely affect the applications using the link. Second, I would not compare iptraf with bing. Iptraf measures what is actually going through your interfaces while bing does a correlation between the RTT and delays for different packet sizes.
    – c-garcia
    Jan 28, 2015 at 20:34
  • at least bing has a different, very interesting approach. It is clear that if it is not possible to measure something by actually using it, then it must be estimated. ATM I am not sure if what bing does is the best available method, but it is a good starting point. Jan 28, 2015 at 21:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .