Bandwidth measurement

[acastilloh5690]

Hello, I have an MPLS connection between sites and I need to measure the bandwidth capacity of the MPLS links. How can I do it? I have Catalyst switch 3750 and 3850.

[balaji.bandi]

if i understand correctly you want to test the link as per the delivery of MPLS

in industry test, we do iperf testing between site to site. - https://iperf.fr/iperf-doc.php

is this what you looking ?

[Joseph W. Doherty]

Depends what exactly you're trying to measure.

I presume your MPLS vendor should be able to tell you what bandwidth they are providing you. Is there some reason you believe you're not obtaining the bandwidth they say they are providing?

[acastilloh5690]

You are right, my ISP send me a report with the bandwidth provided, but in many opportunities we have noticed that they have reduced the bandwidth that we have by contract, so I need a way to verify that offently. Is there any way?

[Joseph W. Doherty]

The basic bandwidth test tools, like iperf, are a good way to perform some basic bandwidth checks, but there are lots of issues when dealing with obtainable bandwidth. For example, if your MPLS provider is policing bandwidth, policers often drop packets during bursts. I.e. your average obtainable bandwidth, for traffic like TCP, will then be less what can be obtained by UDP at a set rate.

[Giuseppe Larosa]

Hello acastilloh5690,

the suggestion to use iperf is a good one.

However, to test bandwidth you should use UDP streams, because TCP throughput is influenced by the bandwidth delay product. If the two sites are geographically far TCP results will be lower then UDP ones even if the service provider is doing its best and it is conforming to contracted SLAs.

To be noted in some cases you can send multiple TCP sessions concurrently over the path and the aggregate throughput will be more close to single UDP session test results.

Hope to help

Giuseppe

[acastilloh5690]

How can I do it?

[Giuseppe Larosa]

Hello acastilloh5690,

you need to install the iperf SW on two different PCs one located in site1 and one located in site2.

This is the way to use a tool like iperf.

One PC will act as client and the other will act as server and you can perform different tests based on TCP or UDP.

You can make multiple tests and you can exchange roles over time. To see also if there is any change in performance when changing the server role position.

Note:

This kind of tests are invasive and will have impact on normal user traffic so tests should be performed when users are not in office.

Hope to help

Giuseppe

[acastilloh5690]

Thank you.

[Joseph W. Doherty]

"Note:

This kind of tests are invasive and will have impact on normal user traffic so tests should be performed when users are not in office."

BTW, if you have a QoS setup that allows you to de-prioritize the bandwidth test traffic relative to all other traffic, the impact can be minimal.

[balaji.bandi]

This what i have suggested on other post (use iperf), and suggest to do this test due to impact and clear outcome, suggest to do it in off peak business hours, so you have base line zero business traffic to test different methods.

[Joseph W. Doherty]

I agree with Giuseppe that UDP often makes it easier to test bandwidth capacity as TCP does have issues (like BDP as mentioned [along with others] that can impact is bandwidth utilization, although those issues can often be what is slowing your "normal" traffic). (Often an easy UDP test is to send a bit more than the bandwidth guaranteed rate and see if the other side receives the guaranteed rate.)